6abb410 fix(provider): correct Copilot routing and model availability · zut

M docs/providers.md +49 -0


@@ -90,6 +90,55 @@ These providers support subscription login:
90	90		OAuth tokens are stored in `$ZOT_HOME/auth.json` and refreshed when refresh is
91	91		available.
92	92
	93	+	## GitHub Copilot model availability
	94	+
	95	+	The built-in Copilot catalog excludes retired models and utility-only models
	96	+	that are not selectable for chat. The default is `claude-sonnet-5`.
	97	+	Availability still depends on your plan and organization policies. At startup,
	98	+	zot queries Copilot's authenticated `/models` endpoint in the background and
	99	+	filters the runtime catalog using picker visibility, policy state, and tool-call
	100	+	support. Individual accounts with no eligible picker entries can fall back to
	101	+	explicitly enabled model policies; other account types cannot. zot never enables
	102	+	policies automatically.
	103	+
	104	+	Account availability stays in memory, not the shared model cache. Until discovery
	105	+	succeeds, the static catalog remains visible; failed refreshes retain the previous
	106	+	snapshot. Command-backed credentials are not executed for background discovery.
	107	+	Restart zot after changing accounts to refresh the snapshot. This filter is a
	108	+	selection aid, not an authorization boundary. The website displays the public
	109	+	catalog, not any individual account's availability.
	110	+
	111	+	Claude uses the Anthropic Messages API with Copilot Bearer authentication, without
	112	+	Claude Code identity injection. GPT, Grok, and MAI use Responses. Gemini and Kimi
	113	+	use Chat Completions without `reasoning_effort`; their reasoning is managed by the
	114	+	endpoint. All three routes send user/agent initiator metadata and image-request
	115	+	headers where appropriate.
	116	+
	117	+	As of September 8, 2026, the catalog includes 27 public chat models: Claude
	118	+	Fable 5/5.1, Haiku 4.5, Opus 4.7/4.8/4.8 Fast/5, Sonnet 5, Gemini
	119	+	3.5/3.6/3.7/3.8 Flash, GPT-5 mini, GPT-5.3-Codex, GPT-5.4/5.4 mini/5.5,
	120	+	GPT-5.6 Luna/Sol/Terra, GPT-6 Astra, Grok 4.5/4.6, Kimi K2.7 Code/K3,
	121	+	and MAI-Code-1-Flash/1.1-Flash. MAI-Code-1-Flash uses the public picker ID
	122	+	`mai-code-1-flash-picker` and is scheduled to retire on September 10, 2026.
	123	+
	124	+	Internal search, execution, and compaction models are not chat choices.
	125	+	GPT-5.4 nano is excluded because GitHub offers it in the Codex VS Code
	126	+	extension, not Copilot Chat. New entries use Copilot-specific limits and token
	127	+	prices from [models.dev](https://models.dev); these are not a quote for your
	128	+	subscription bill. Existing zero-priced entries do not imply unlimited free use.
	129	+
	130	+	Claude Fable models have different data-retention terms: Anthropic retains
	131	+	prompts and outputs by default for safety classifiers. Eligible enterprises
	132	+	can request a time-limited zero-data-retention exception from GitHub; enabling
	133	+	a model does not itself grant that exception. Review
	134	+	[GitHub's model availability and data-retention notes](https://docs.github.com/en/copilot/reference/ai-models/supported-models)
	135	+	before using Fable with sensitive data.
	136	+
	137	+	Sonnet 4.6 is also excluded from the general catalog: GitHub retired it on
	138	+	September 1, 2026, except for individual annual Copilot Pro and Pro+ subscribers.
	139	+	Eligible subscribers can add it through `$ZOT_HOME/models.json`.
	140	+	See [GitHub's model retirement history and plan exceptions](https://docs.github.com/en/copilot/reference/ai-models/supported-models#model-retirement-history).
	141	+
93	142		## GPT-6 Astra
94	143
95	144		Select `gpt-6-astra` with provider `openai`, `openai-responses`,


@@ -90,6 +90,55 @@ These providers support subscription login:
90	OAuth tokens are stored in `$ZOT_HOME/auth.json` and refreshed when refresh is	90		OAuth tokens are stored in `$ZOT_HOME/auth.json` and refreshed when refresh is
91	available.	91		available.
92		92
		93	+	## GitHub Copilot model availability
		94	+
		95	+	The built-in Copilot catalog excludes retired models and utility-only models
		96	+	that are not selectable for chat. The default is `claude-sonnet-5`.
		97	+	Availability still depends on your plan and organization policies. At startup,
		98	+	zot queries Copilot's authenticated `/models` endpoint in the background and
		99	+	filters the runtime catalog using picker visibility, policy state, and tool-call
		100	+	support. Individual accounts with no eligible picker entries can fall back to
		101	+	explicitly enabled model policies; other account types cannot. zot never enables
		102	+	policies automatically.
		103	+
		104	+	Account availability stays in memory, not the shared model cache. Until discovery
		105	+	succeeds, the static catalog remains visible; failed refreshes retain the previous
		106	+	snapshot. Command-backed credentials are not executed for background discovery.
		107	+	Restart zot after changing accounts to refresh the snapshot. This filter is a
		108	+	selection aid, not an authorization boundary. The website displays the public
		109	+	catalog, not any individual account's availability.
		110	+
		111	+	Claude uses the Anthropic Messages API with Copilot Bearer authentication, without
		112	+	Claude Code identity injection. GPT, Grok, and MAI use Responses. Gemini and Kimi
		113	+	use Chat Completions without `reasoning_effort`; their reasoning is managed by the
		114	+	endpoint. All three routes send user/agent initiator metadata and image-request
		115	+	headers where appropriate.
		116	+
		117	+	As of September 8, 2026, the catalog includes 27 public chat models: Claude
		118	+	Fable 5/5.1, Haiku 4.5, Opus 4.7/4.8/4.8 Fast/5, Sonnet 5, Gemini
		119	+	3.5/3.6/3.7/3.8 Flash, GPT-5 mini, GPT-5.3-Codex, GPT-5.4/5.4 mini/5.5,
		120	+	GPT-5.6 Luna/Sol/Terra, GPT-6 Astra, Grok 4.5/4.6, Kimi K2.7 Code/K3,
		121	+	and MAI-Code-1-Flash/1.1-Flash. MAI-Code-1-Flash uses the public picker ID
		122	+	`mai-code-1-flash-picker` and is scheduled to retire on September 10, 2026.
		123	+
		124	+	Internal search, execution, and compaction models are not chat choices.
		125	+	GPT-5.4 nano is excluded because GitHub offers it in the Codex VS Code
		126	+	extension, not Copilot Chat. New entries use Copilot-specific limits and token
		127	+	prices from [models.dev](https://models.dev); these are not a quote for your
		128	+	subscription bill. Existing zero-priced entries do not imply unlimited free use.
		129	+
		130	+	Claude Fable models have different data-retention terms: Anthropic retains
		131	+	prompts and outputs by default for safety classifiers. Eligible enterprises
		132	+	can request a time-limited zero-data-retention exception from GitHub; enabling
		133	+	a model does not itself grant that exception. Review
		134	+	[GitHub's model availability and data-retention notes](https://docs.github.com/en/copilot/reference/ai-models/supported-models)
		135	+	before using Fable with sensitive data.
		136	+
		137	+	Sonnet 4.6 is also excluded from the general catalog: GitHub retired it on
		138	+	September 1, 2026, except for individual annual Copilot Pro and Pro+ subscribers.
		139	+	Eligible subscribers can add it through `$ZOT_HOME/models.json`.
		140	+	See [GitHub's model retirement history and plan exceptions](https://docs.github.com/en/copilot/reference/ai-models/supported-models#model-retirement-history).
		141	+
93	## GPT-6 Astra	142		## GPT-6 Astra
94		143
95	Select `gpt-6-astra` with provider `openai`, `openai-responses`,	144		Select `gpt-6-astra` with provider `openai`, `openai-responses`,

M packages/agent/build.go +1 -1


@@ -215,7 +215,7 @@ func defaultModelForProvider(prov string) string {
215	215		case "azure-openai-responses":
216	216		return "gpt-5"
217	217		case "github-copilot":
218		-	return "claude-sonnet-4.5"
	218	+	return "claude-sonnet-5"
219	219		default:
220	220		// Custom providers: pick the first model from the catalog for
221	221		// that provider, or fall back to the global default.


@@ -215,7 +215,7 @@ func defaultModelForProvider(prov string) string {
215		case "azure-openai-responses":	215		case "azure-openai-responses":
216		return "gpt-5"	216		return "gpt-5"
217		case "github-copilot":	217		case "github-copilot":
218	-	return "claude-sonnet-4.5"	218	+	return "claude-sonnet-5"
219		default:	219		default:
220		// Custom providers: pick the first model from the catalog for	220		// Custom providers: pick the first model from the catalog for
221		// that provider, or fall back to the global default.	221		// that provider, or fall back to the global default.

A packages/agent/copilot_availability_test.go +19 -0

@@ -0,0 +1,19 @@
	1	+	package agent
	2	+
	3	+	import (
	4	+	"testing"
	5	+
	6	+	"github.com/patriceckhart/zot/packages/provider"
	7	+	)
	8	+
	9	+	func TestRefreshCopilotAvailabilityWithoutCredentials(t *testing.T) {
	10	+	t.Setenv("ZOT_HOME", t.TempDir())
	11	+	t.Setenv("COPILOT_GITHUB_TOKEN", "")
	12	+	t.Setenv("GITHUB_COPILOT_TOKEN", "")
	13	+	provider.SetModelAvailability("github-copilot", []string{})
	14	+	t.Cleanup(func() { provider.SetModelAvailability("github-copilot", nil) })
	15	+	refreshCopilotModelAvailability()
	16	+	if len(provider.ModelsForProvider("github-copilot")) == 0 {
	17	+	t.Fatal("account restriction was retained after credentials disappeared")
	18	+	}
	19	+	}

@@ -0,0 +1,19 @@
	1	+	package agent
	2	+
	3	+	import (
	4	+	"testing"
	5	+
	6	+	"github.com/patriceckhart/zot/packages/provider"
	7	+	)
	8	+
	9	+	func TestRefreshCopilotAvailabilityWithoutCredentials(t *testing.T) {
	10	+	t.Setenv("ZOT_HOME", t.TempDir())
	11	+	t.Setenv("COPILOT_GITHUB_TOKEN", "")
	12	+	t.Setenv("GITHUB_COPILOT_TOKEN", "")
	13	+	provider.SetModelAvailability("github-copilot", []string{})
	14	+	t.Cleanup(func() { provider.SetModelAvailability("github-copilot", nil) })
	15	+	refreshCopilotModelAvailability()
	16	+	if len(provider.ModelsForProvider("github-copilot")) == 0 {
	17	+	t.Fatal("account restriction was retained after credentials disappeared")
	18	+	}
	19	+	}

A packages/agent/copilot_default_test.go +17 -0

@@ -0,0 +1,17 @@
	1	+	package agent
	2	+
	3	+	import (
	4	+	"testing"
	5	+
	6	+	"github.com/patriceckhart/zot/packages/provider"
	7	+	)
	8	+
	9	+	func TestCopilotDefaultUsesSonnet5(t *testing.T) {
	10	+	id := defaultModelForProvider("github-copilot")
	11	+	if id != "claude-sonnet-5" {
	12	+	t.Fatalf("Copilot default = %q, want claude-sonnet-5", id)
	13	+	}
	14	+	if _, err := provider.FindModel("github-copilot", id); err != nil {
	15	+	t.Fatalf("Copilot default missing from catalog: %v", err)
	16	+	}
	17	+	}

@@ -0,0 +1,17 @@
	1	+	package agent
	2	+
	3	+	import (
	4	+	"testing"
	5	+
	6	+	"github.com/patriceckhart/zot/packages/provider"
	7	+	)
	8	+
	9	+	func TestCopilotDefaultUsesSonnet5(t *testing.T) {
	10	+	id := defaultModelForProvider("github-copilot")
	11	+	if id != "claude-sonnet-5" {
	12	+	t.Fatalf("Copilot default = %q, want claude-sonnet-5", id)
	13	+	}
	14	+	if _, err := provider.FindModel("github-copilot", id); err != nil {
	15	+	t.Fatalf("Copilot default missing from catalog: %v", err)
	16	+	}
	17	+	}

M packages/agent/modelsync.go +17 -0


@@ -141,6 +141,7 @@ func ValidateAndRepairConfig() {
141	141		// the baked-in catalog if this fails.
142	142		func RefreshModelsAsync() {
143	143		go refreshModels()
	144	+	go refreshCopilotModelAvailability()
144	145		go func() {
145	146		ctx, cancel := context.WithTimeout(context.Background(), 3*time.Second)
146	147		defer cancel()

@@ -148,6 +149,22 @@ func RefreshModelsAsync() {
148	149		}()
149	150		}
150	151
	152	+	// Copilot availability is account-specific and must not use the shared disk cache.
	153	+	func refreshCopilotModelAvailability() {
	154	+	ctx, cancel := context.WithTimeout(context.Background(), 20*time.Second)
	155	+	defer cancel()
	156	+	cred, _, err := resolveCredentialForBackground(ctx, "github-copilot")
	157	+	if err != nil {
	158	+	provider.SetModelAvailability("github-copilot", nil)
	159	+	return
	160	+	}
	161	+	ids, err := provider.DiscoverCopilotAvailableModels(ctx, cred)
	162	+	if err != nil {
	163	+	return // Keep the previous snapshot, or the catalog on first-load failure.
	164	+	}
	165	+	provider.SetModelAvailability("github-copilot", ids)
	166	+	}
	167	+
151	168		// RefreshLlamaCPPModels adds the router's currently loaded models to the
152	169		// active catalog. Unloaded models remain in the management UI and cannot be
153	170		// selected for inference until they are loaded.


@@ -141,6 +141,7 @@ func ValidateAndRepairConfig() {
141	// the baked-in catalog if this fails.	141		// the baked-in catalog if this fails.
142	func RefreshModelsAsync() {	142		func RefreshModelsAsync() {
143	go refreshModels()	143		go refreshModels()
		144	+	go refreshCopilotModelAvailability()
144	go func() {	145		go func() {
145	ctx, cancel := context.WithTimeout(context.Background(), 3*time.Second)	146		ctx, cancel := context.WithTimeout(context.Background(), 3*time.Second)
146	defer cancel()	147		defer cancel()

@@ -148,6 +149,22 @@ func RefreshModelsAsync() {
148	}()	149		}()
149	}	150		}
150		151
		152	+	// Copilot availability is account-specific and must not use the shared disk cache.
		153	+	func refreshCopilotModelAvailability() {
		154	+	ctx, cancel := context.WithTimeout(context.Background(), 20*time.Second)
		155	+	defer cancel()
		156	+	cred, _, err := resolveCredentialForBackground(ctx, "github-copilot")
		157	+	if err != nil {
		158	+	provider.SetModelAvailability("github-copilot", nil)
		159	+	return
		160	+	}
		161	+	ids, err := provider.DiscoverCopilotAvailableModels(ctx, cred)
		162	+	if err != nil {
		163	+	return // Keep the previous snapshot, or the catalog on first-load failure.
		164	+	}
		165	+	provider.SetModelAvailability("github-copilot", ids)
		166	+	}
		167	+
151	// RefreshLlamaCPPModels adds the router's currently loaded models to the	168		// RefreshLlamaCPPModels adds the router's currently loaded models to the
152	// active catalog. Unloaded models remain in the management UI and cannot be	169		// active catalog. Unloaded models remain in the management UI and cannot be
153	// selected for inference until they are loaded.	170		// selected for inference until they are loaded.

M packages/provider/catalog_builtin.go +25 -13


@@ -12,7 +12,14 @@ package provider
12	12		// the small seed list for kimi / deepseek / google) are not duplicated
13	13		// here; they take precedence on (provider, id) match.
14	14
15		-	func init() { Catalog = append(Catalog, builtinCatalog...) }
	15	+	func init() {
	16	+	for i := range builtinCatalog {
	17	+	if builtinCatalog[i].Provider == "github-copilot" {
	18	+	configureCopilotModel(&builtinCatalog[i])
	19	+	}
	20	+	}
	21	+	Catalog = append(Catalog, builtinCatalog...)
	22	+	}
16	23
17	24		var builtinCatalog = []Model{
18	25		// ----- amazon-bedrock -----

@@ -139,23 +146,29 @@ var builtinCatalog = []Model{
139	146		{Provider: "fireworks", ID: "accounts/fireworks/routers/glm-5p1-fast", DisplayName: "GLM 5.1 Fast", ContextWindow: 202800, MaxOutput: 131072, Reasoning: true, PriceInput: 2.8, PriceOutput: 8.8, PriceCacheRead: 0.52, BaseURL: "https://api.fireworks.ai/inference"},
140	147		{Provider: "fireworks", ID: "accounts/fireworks/routers/kimi-k2p6-turbo", DisplayName: "Kimi K2.6 Turbo", ContextWindow: 262000, MaxOutput: 262000, Reasoning: true, PriceInput: 2, PriceOutput: 8, PriceCacheRead: 0.3, BaseURL: "https://api.fireworks.ai/inference"},
141	148		// ----- github-copilot -----
	149	+	// Public chat models checked against GitHub's supported-model list on 2026-09-08.
	150	+	// New model limits and token prices: https://models.dev/api.json (github-copilot).
	151	+	// Copilot's fast Opus model ID comes from the inference API's available-model list.
	152	+	{Provider: "github-copilot", ID: "claude-fable-5", DisplayName: "Claude Fable 5", ContextWindow: 1000000, MaxOutput: 128000, Reasoning: true, AdaptiveThinking: true, PriceInput: 10, PriceOutput: 50, PriceCacheRead: 1, PriceCacheWrite: 12.5, BaseURL: "https://api.individual.githubcopilot.com"},
	153	+	{Provider: "github-copilot", ID: "claude-fable-5.1", DisplayName: "Claude Fable 5.1", ContextWindow: 1000000, MaxOutput: 128000, Reasoning: true, AdaptiveThinking: true, PriceInput: 10, PriceOutput: 50, PriceCacheRead: 0.25, PriceCacheWrite: 12.5, BaseURL: "https://api.individual.githubcopilot.com"},
	154	+	{Provider: "github-copilot", ID: "claude-opus-4.8-fast", DisplayName: "Claude Opus 4.8 (fast mode)", ContextWindow: 200000, MaxOutput: 64000, Reasoning: true, AdaptiveThinking: true, PriceInput: 10, PriceOutput: 50, PriceCacheRead: 1, PriceCacheWrite: 12.5, BaseURL: "https://api.individual.githubcopilot.com"},
	155	+	{Provider: "github-copilot", ID: "claude-opus-5", DisplayName: "Claude Opus 5", ContextWindow: 1000000, MaxOutput: 64000, Reasoning: true, AdaptiveThinking: true, PriceInput: 5, PriceOutput: 25, PriceCacheRead: 0.5, PriceCacheWrite: 6.25, BaseURL: "https://api.individual.githubcopilot.com"},
	156	+	{Provider: "github-copilot", ID: "gemini-3.6-flash", DisplayName: "Gemini 3.6 Flash", ContextWindow: 1000000, MaxOutput: 64000, Reasoning: true, PriceInput: 0.75, PriceOutput: 3.75, PriceCacheRead: 0.075, BaseURL: "https://api.individual.githubcopilot.com"},
	157	+	{Provider: "github-copilot", ID: "gemini-3.7-flash", DisplayName: "Gemini 3.7 Flash", ContextWindow: 1000000, MaxOutput: 64000, Reasoning: true, PriceInput: 0.75, PriceOutput: 3.75, PriceCacheRead: 0.075, BaseURL: "https://api.individual.githubcopilot.com"},
	158	+	{Provider: "github-copilot", ID: "gemini-3.8-flash", DisplayName: "Gemini 3.8 Flash", ContextWindow: 1000000, MaxOutput: 64000, Reasoning: true, PriceInput: 0.75, PriceOutput: 3.75, PriceCacheRead: 0.075, BaseURL: "https://api.individual.githubcopilot.com"},
	159	+	{Provider: "github-copilot", ID: "grok-4.5", DisplayName: "Grok 4.5", ContextWindow: 500000, MaxOutput: 128000, Reasoning: true, PriceInput: 2, PriceOutput: 6, PriceCacheRead: 0.5, PriceTierInputTokens: 200000, PriceInputAbove: 4, PriceOutputAbove: 12, PriceCacheReadAbove: 1, BaseURL: "https://api.individual.githubcopilot.com"},
	160	+	{Provider: "github-copilot", ID: "grok-4.6", DisplayName: "Grok 4.6", ContextWindow: 500000, MaxOutput: 128000, Reasoning: true, PriceInput: 2, PriceOutput: 6, PriceCacheRead: 0.5, PriceTierInputTokens: 200000, PriceInputAbove: 4, PriceOutputAbove: 12, PriceCacheReadAbove: 1, BaseURL: "https://api.individual.githubcopilot.com"},
	161	+	{Provider: "github-copilot", ID: "kimi-k2.7-code", DisplayName: "Kimi K2.7 Code", ContextWindow: 256000, MaxOutput: 32000, Reasoning: true, PriceInput: 0.95, PriceOutput: 4, PriceCacheRead: 0.19, BaseURL: "https://api.individual.githubcopilot.com"},
	162	+	{Provider: "github-copilot", ID: "kimi-k3", DisplayName: "Kimi K3", ContextWindow: 1048576, MaxOutput: 131072, Reasoning: true, PriceInput: 3, PriceOutput: 15, PriceCacheRead: 0.3, BaseURL: "https://api.individual.githubcopilot.com"},
	163	+	// MAI-Code-1-Flash remains offered until its scheduled retirement on 2026-09-10.
	164	+	{Provider: "github-copilot", ID: "mai-code-1-flash-picker", DisplayName: "MAI-Code-1-Flash", ContextWindow: 256000, MaxOutput: 128000, Reasoning: true, PriceInput: 0.75, PriceOutput: 4.5, PriceCacheRead: 0.075, BaseURL: "https://api.individual.githubcopilot.com"},
	165	+	{Provider: "github-copilot", ID: "mai-code-1.1-flash", DisplayName: "MAI-Code-1.1-Flash", ContextWindow: 256000, MaxOutput: 128000, Reasoning: true, PriceInput: 0.2, PriceOutput: 1.2, PriceCacheRead: 0.02, BaseURL: "https://api.individual.githubcopilot.com"},
142	166		{Provider: "github-copilot", ID: "claude-haiku-4.5", DisplayName: "Claude Haiku 4.5", ContextWindow: 144000, MaxOutput: 32000, Reasoning: true, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://api.individual.githubcopilot.com"},
143		-	{Provider: "github-copilot", ID: "claude-opus-4.5", DisplayName: "Claude Opus 4.5", ContextWindow: 160000, MaxOutput: 32000, Reasoning: true, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://api.individual.githubcopilot.com"},
144		-	{Provider: "github-copilot", ID: "claude-opus-4.6", DisplayName: "Claude Opus 4.6", ContextWindow: 1000000, MaxOutput: 64000, Reasoning: true, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://api.individual.githubcopilot.com"},
145	167		{Provider: "github-copilot", ID: "claude-opus-4.7", DisplayName: "Claude Opus 4.7", ContextWindow: 144000, MaxOutput: 64000, Reasoning: true, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://api.individual.githubcopilot.com"},
146	168		{Provider: "github-copilot", ID: "claude-opus-4.8", DisplayName: "Claude Opus 4.8", ContextWindow: 144000, MaxOutput: 64000, Reasoning: true, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://api.individual.githubcopilot.com"},
147		-	{Provider: "github-copilot", ID: "claude-sonnet-4.5", DisplayName: "Claude Sonnet 4.5", ContextWindow: 144000, MaxOutput: 32000, Reasoning: true, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://api.individual.githubcopilot.com"},
148		-	{Provider: "github-copilot", ID: "claude-sonnet-4.6", DisplayName: "Claude Sonnet 4.6", ContextWindow: 1000000, MaxOutput: 32000, Reasoning: true, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://api.individual.githubcopilot.com"},
149	169		{Provider: "github-copilot", ID: "claude-sonnet-5", DisplayName: "Claude Sonnet 5", ContextWindow: 1000000, MaxOutput: 128000, Reasoning: true, AdaptiveThinking: true, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://api.individual.githubcopilot.com"},
150		-	{Provider: "github-copilot", ID: "gemini-2.5-pro", DisplayName: "Gemini 2.5 Pro", ContextWindow: 128000, MaxOutput: 64000, Reasoning: false, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://api.individual.githubcopilot.com"},
151		-	{Provider: "github-copilot", ID: "gemini-3-flash-preview", DisplayName: "Gemini 3 Flash", ContextWindow: 128000, MaxOutput: 64000, Reasoning: true, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://api.individual.githubcopilot.com"},
152		-	{Provider: "github-copilot", ID: "gemini-3.1-pro-preview", DisplayName: "Gemini 3.1 Pro Preview", ContextWindow: 128000, MaxOutput: 64000, Reasoning: true, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://api.individual.githubcopilot.com"},
153	170		{Provider: "github-copilot", ID: "gemini-3.5-flash", DisplayName: "Gemini 3.5 Flash", ContextWindow: 128000, MaxOutput: 64000, Reasoning: true, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://api.individual.githubcopilot.com"},
154		-	{Provider: "github-copilot", ID: "gpt-4.1", DisplayName: "GPT-4.1", ContextWindow: 128000, MaxOutput: 16384, Reasoning: false, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://api.individual.githubcopilot.com"},
155		-	{Provider: "github-copilot", ID: "gpt-4o", DisplayName: "GPT-4o", ContextWindow: 128000, MaxOutput: 4096, Reasoning: false, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://api.individual.githubcopilot.com"},
156	171		{Provider: "github-copilot", ID: "gpt-5-mini", DisplayName: "GPT-5-mini", ContextWindow: 264000, MaxOutput: 64000, Reasoning: true, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://api.individual.githubcopilot.com"},
157		-	{Provider: "github-copilot", ID: "gpt-5.2", DisplayName: "GPT-5.2", ContextWindow: 264000, MaxOutput: 64000, Reasoning: true, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://api.individual.githubcopilot.com"},
158		-	{Provider: "github-copilot", ID: "gpt-5.2-codex", DisplayName: "GPT-5.2-Codex", ContextWindow: 400000, MaxOutput: 128000, Reasoning: true, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://api.individual.githubcopilot.com"},
159	172		{Provider: "github-copilot", ID: "gpt-5.3-codex", DisplayName: "GPT-5.3-Codex", ContextWindow: 400000, MaxOutput: 128000, Reasoning: true, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://api.individual.githubcopilot.com"},
160	173		{Provider: "github-copilot", ID: "gpt-5.4", DisplayName: "GPT-5.4", ContextWindow: 400000, MaxOutput: 128000, Reasoning: true, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://api.individual.githubcopilot.com"},
161	174		{Provider: "github-copilot", ID: "gpt-5.4-mini", DisplayName: "GPT-5.4 Mini", ContextWindow: 400000, MaxOutput: 128000, Reasoning: true, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://api.individual.githubcopilot.com"},

@@ -164,7 +177,6 @@ var builtinCatalog = []Model{
164	177		{Provider: "github-copilot", ID: "gpt-5.6-terra", DisplayName: "GPT-5.6 Terra", API: APIResponses, ContextWindow: 400000, MaxOutput: 128000, Reasoning: true, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://api.individual.githubcopilot.com"},
165	178		{Provider: "github-copilot", ID: "gpt-5.6-luna", DisplayName: "GPT-5.6 Luna", API: APIResponses, ContextWindow: 400000, MaxOutput: 128000, Reasoning: true, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://api.individual.githubcopilot.com"},
166	179		{Provider: "github-copilot", ID: "gpt-6-astra", DisplayName: "GPT-6 Astra", API: APIResponses, ContextWindow: 1050000, MaxOutput: 128000, Reasoning: true, PriceInput: 10, PriceOutput: 50, PriceCacheRead: 1, PriceCacheWrite: 12.5, PriceTierInputTokens: 272000, PriceInputAbove: 20, PriceOutputAbove: 75, PriceCacheReadAbove: 2, PriceCacheWriteAbove: 25, BaseURL: "https://api.individual.githubcopilot.com"},
167		-	{Provider: "github-copilot", ID: "grok-code-fast-1", DisplayName: "Grok Code Fast 1", ContextWindow: 128000, MaxOutput: 64000, Reasoning: true, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://api.individual.githubcopilot.com"},
168	180		// ----- google -----
169	181		{Provider: "google", ID: "gemini-3-flash-preview", DisplayName: "Gemini 3 Flash Preview", ContextWindow: 1048576, MaxOutput: 65536, Reasoning: true, PriceInput: 0.5, PriceOutput: 3, PriceCacheRead: 0.05, BaseURL: "https://generativelanguage.googleapis.com/v1beta"},
170	182		{Provider: "google", ID: "gemini-3-pro-preview", DisplayName: "Gemini 3 Pro Preview", ContextWindow: 1048576, MaxOutput: 65536, Reasoning: true, PriceInput: 2, PriceOutput: 12, PriceCacheRead: 0.2, BaseURL: "https://generativelanguage.googleapis.com/v1beta"},

 // the small seed list for kimi / deepseek / google) are not duplicated
 // here; they take precedence on (provider, id) match.
-func init() { Catalog = append(Catalog, builtinCatalog...) }
+func init() {
+	for i := range builtinCatalog {
+		if builtinCatalog[i].Provider == "github-copilot" {
+			configureCopilotModel(&builtinCatalog[i])
+		}
+	}
+	Catalog = append(Catalog, builtinCatalog...)
+}
 var builtinCatalog = []Model{
 	// ----- amazon-bedrock -----
 	{Provider: "fireworks", ID: "accounts/fireworks/routers/glm-5p1-fast", DisplayName: "GLM 5.1 Fast", ContextWindow: 202800, MaxOutput: 131072, Reasoning: true, PriceInput: 2.8, PriceOutput: 8.8, PriceCacheRead: 0.52, BaseURL: "https://api.fireworks.ai/inference"},
 	{Provider: "fireworks", ID: "accounts/fireworks/routers/kimi-k2p6-turbo", DisplayName: "Kimi K2.6 Turbo", ContextWindow: 262000, MaxOutput: 262000, Reasoning: true, PriceInput: 2, PriceOutput: 8, PriceCacheRead: 0.3, BaseURL: "https://api.fireworks.ai/inference"},
 	// ----- github-copilot -----
+	// Public chat models checked against GitHub's supported-model list on 2026-09-08.
+	// New model limits and token prices: https://models.dev/api.json (github-copilot).
+	// Copilot's fast Opus model ID comes from the inference API's available-model list.
+	{Provider: "github-copilot", ID: "claude-fable-5", DisplayName: "Claude Fable 5", ContextWindow: 1000000, MaxOutput: 128000, Reasoning: true, AdaptiveThinking: true, PriceInput: 10, PriceOutput: 50, PriceCacheRead: 1, PriceCacheWrite: 12.5, BaseURL: "https://api.individual.githubcopilot.com"},
+	{Provider: "github-copilot", ID: "claude-fable-5.1", DisplayName: "Claude Fable 5.1", ContextWindow: 1000000, MaxOutput: 128000, Reasoning: true, AdaptiveThinking: true, PriceInput: 10, PriceOutput: 50, PriceCacheRead: 0.25, PriceCacheWrite: 12.5, BaseURL: "https://api.individual.githubcopilot.com"},
+	{Provider: "github-copilot", ID: "claude-opus-4.8-fast", DisplayName: "Claude Opus 4.8 (fast mode)", ContextWindow: 200000, MaxOutput: 64000, Reasoning: true, AdaptiveThinking: true, PriceInput: 10, PriceOutput: 50, PriceCacheRead: 1, PriceCacheWrite: 12.5, BaseURL: "https://api.individual.githubcopilot.com"},
+	{Provider: "github-copilot", ID: "claude-opus-5", DisplayName: "Claude Opus 5", ContextWindow: 1000000, MaxOutput: 64000, Reasoning: true, AdaptiveThinking: true, PriceInput: 5, PriceOutput: 25, PriceCacheRead: 0.5, PriceCacheWrite: 6.25, BaseURL: "https://api.individual.githubcopilot.com"},
+	{Provider: "github-copilot", ID: "gemini-3.6-flash", DisplayName: "Gemini 3.6 Flash", ContextWindow: 1000000, MaxOutput: 64000, Reasoning: true, PriceInput: 0.75, PriceOutput: 3.75, PriceCacheRead: 0.075, BaseURL: "https://api.individual.githubcopilot.com"},
+	{Provider: "github-copilot", ID: "gemini-3.7-flash", DisplayName: "Gemini 3.7 Flash", ContextWindow: 1000000, MaxOutput: 64000, Reasoning: true, PriceInput: 0.75, PriceOutput: 3.75, PriceCacheRead: 0.075, BaseURL: "https://api.individual.githubcopilot.com"},
+	{Provider: "github-copilot", ID: "gemini-3.8-flash", DisplayName: "Gemini 3.8 Flash", ContextWindow: 1000000, MaxOutput: 64000, Reasoning: true, PriceInput: 0.75, PriceOutput: 3.75, PriceCacheRead: 0.075, BaseURL: "https://api.individual.githubcopilot.com"},
+	{Provider: "github-copilot", ID: "grok-4.5", DisplayName: "Grok 4.5", ContextWindow: 500000, MaxOutput: 128000, Reasoning: true, PriceInput: 2, PriceOutput: 6, PriceCacheRead: 0.5, PriceTierInputTokens: 200000, PriceInputAbove: 4, PriceOutputAbove: 12, PriceCacheReadAbove: 1, BaseURL: "https://api.individual.githubcopilot.com"},
+	{Provider: "github-copilot", ID: "grok-4.6", DisplayName: "Grok 4.6", ContextWindow: 500000, MaxOutput: 128000, Reasoning: true, PriceInput: 2, PriceOutput: 6, PriceCacheRead: 0.5, PriceTierInputTokens: 200000, PriceInputAbove: 4, PriceOutputAbove: 12, PriceCacheReadAbove: 1, BaseURL: "https://api.individual.githubcopilot.com"},
+	{Provider: "github-copilot", ID: "kimi-k2.7-code", DisplayName: "Kimi K2.7 Code", ContextWindow: 256000, MaxOutput: 32000, Reasoning: true, PriceInput: 0.95, PriceOutput: 4, PriceCacheRead: 0.19, BaseURL: "https://api.individual.githubcopilot.com"},
+	{Provider: "github-copilot", ID: "kimi-k3", DisplayName: "Kimi K3", ContextWindow: 1048576, MaxOutput: 131072, Reasoning: true, PriceInput: 3, PriceOutput: 15, PriceCacheRead: 0.3, BaseURL: "https://api.individual.githubcopilot.com"},
+	// MAI-Code-1-Flash remains offered until its scheduled retirement on 2026-09-10.
+	{Provider: "github-copilot", ID: "mai-code-1-flash-picker", DisplayName: "MAI-Code-1-Flash", ContextWindow: 256000, MaxOutput: 128000, Reasoning: true, PriceInput: 0.75, PriceOutput: 4.5, PriceCacheRead: 0.075, BaseURL: "https://api.individual.githubcopilot.com"},
+	{Provider: "github-copilot", ID: "mai-code-1.1-flash", DisplayName: "MAI-Code-1.1-Flash", ContextWindow: 256000, MaxOutput: 128000, Reasoning: true, PriceInput: 0.2, PriceOutput: 1.2, PriceCacheRead: 0.02, BaseURL: "https://api.individual.githubcopilot.com"},
 	{Provider: "github-copilot", ID: "claude-haiku-4.5", DisplayName: "Claude Haiku 4.5", ContextWindow: 144000, MaxOutput: 32000, Reasoning: true, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://api.individual.githubcopilot.com"},
-	{Provider: "github-copilot", ID: "claude-opus-4.5", DisplayName: "Claude Opus 4.5", ContextWindow: 160000, MaxOutput: 32000, Reasoning: true, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://api.individual.githubcopilot.com"},
-	{Provider: "github-copilot", ID: "claude-opus-4.6", DisplayName: "Claude Opus 4.6", ContextWindow: 1000000, MaxOutput: 64000, Reasoning: true, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://api.individual.githubcopilot.com"},
 	{Provider: "github-copilot", ID: "claude-opus-4.7", DisplayName: "Claude Opus 4.7", ContextWindow: 144000, MaxOutput: 64000, Reasoning: true, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://api.individual.githubcopilot.com"},
 	{Provider: "github-copilot", ID: "claude-opus-4.8", DisplayName: "Claude Opus 4.8", ContextWindow: 144000, MaxOutput: 64000, Reasoning: true, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://api.individual.githubcopilot.com"},
-	{Provider: "github-copilot", ID: "claude-sonnet-4.5", DisplayName: "Claude Sonnet 4.5", ContextWindow: 144000, MaxOutput: 32000, Reasoning: true, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://api.individual.githubcopilot.com"},
-	{Provider: "github-copilot", ID: "claude-sonnet-4.6", DisplayName: "Claude Sonnet 4.6", ContextWindow: 1000000, MaxOutput: 32000, Reasoning: true, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://api.individual.githubcopilot.com"},
 	{Provider: "github-copilot", ID: "claude-sonnet-5", DisplayName: "Claude Sonnet 5", ContextWindow: 1000000, MaxOutput: 128000, Reasoning: true, AdaptiveThinking: true, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://api.individual.githubcopilot.com"},
-	{Provider: "github-copilot", ID: "gemini-2.5-pro", DisplayName: "Gemini 2.5 Pro", ContextWindow: 128000, MaxOutput: 64000, Reasoning: false, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://api.individual.githubcopilot.com"},
-	{Provider: "github-copilot", ID: "gemini-3-flash-preview", DisplayName: "Gemini 3 Flash", ContextWindow: 128000, MaxOutput: 64000, Reasoning: true, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://api.individual.githubcopilot.com"},
-	{Provider: "github-copilot", ID: "gemini-3.1-pro-preview", DisplayName: "Gemini 3.1 Pro Preview", ContextWindow: 128000, MaxOutput: 64000, Reasoning: true, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://api.individual.githubcopilot.com"},
 	{Provider: "github-copilot", ID: "gemini-3.5-flash", DisplayName: "Gemini 3.5 Flash", ContextWindow: 128000, MaxOutput: 64000, Reasoning: true, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://api.individual.githubcopilot.com"},
-	{Provider: "github-copilot", ID: "gpt-4.1", DisplayName: "GPT-4.1", ContextWindow: 128000, MaxOutput: 16384, Reasoning: false, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://api.individual.githubcopilot.com"},
-	{Provider: "github-copilot", ID: "gpt-4o", DisplayName: "GPT-4o", ContextWindow: 128000, MaxOutput: 4096, Reasoning: false, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://api.individual.githubcopilot.com"},
 	{Provider: "github-copilot", ID: "gpt-5-mini", DisplayName: "GPT-5-mini", ContextWindow: 264000, MaxOutput: 64000, Reasoning: true, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://api.individual.githubcopilot.com"},
-	{Provider: "github-copilot", ID: "gpt-5.2", DisplayName: "GPT-5.2", ContextWindow: 264000, MaxOutput: 64000, Reasoning: true, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://api.individual.githubcopilot.com"},
-	{Provider: "github-copilot", ID: "gpt-5.2-codex", DisplayName: "GPT-5.2-Codex", ContextWindow: 400000, MaxOutput: 128000, Reasoning: true, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://api.individual.githubcopilot.com"},
 	{Provider: "github-copilot", ID: "gpt-5.3-codex", DisplayName: "GPT-5.3-Codex", ContextWindow: 400000, MaxOutput: 128000, Reasoning: true, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://api.individual.githubcopilot.com"},
 	{Provider: "github-copilot", ID: "gpt-5.4", DisplayName: "GPT-5.4", ContextWindow: 400000, MaxOutput: 128000, Reasoning: true, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://api.individual.githubcopilot.com"},
 	{Provider: "github-copilot", ID: "gpt-5.4-mini", DisplayName: "GPT-5.4 Mini", ContextWindow: 400000, MaxOutput: 128000, Reasoning: true, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://api.individual.githubcopilot.com"},
 	{Provider: "github-copilot", ID: "gpt-5.6-terra", DisplayName: "GPT-5.6 Terra", API: APIResponses, ContextWindow: 400000, MaxOutput: 128000, Reasoning: true, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://api.individual.githubcopilot.com"},
 	{Provider: "github-copilot", ID: "gpt-5.6-luna", DisplayName: "GPT-5.6 Luna", API: APIResponses, ContextWindow: 400000, MaxOutput: 128000, Reasoning: true, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://api.individual.githubcopilot.com"},
 	{Provider: "github-copilot", ID: "gpt-6-astra", DisplayName: "GPT-6 Astra", API: APIResponses, ContextWindow: 1050000, MaxOutput: 128000, Reasoning: true, PriceInput: 10, PriceOutput: 50, PriceCacheRead: 1, PriceCacheWrite: 12.5, PriceTierInputTokens: 272000, PriceInputAbove: 20, PriceOutputAbove: 75, PriceCacheReadAbove: 2, PriceCacheWriteAbove: 25, BaseURL: "https://api.individual.githubcopilot.com"},
-	{Provider: "github-copilot", ID: "grok-code-fast-1", DisplayName: "Grok Code Fast 1", ContextWindow: 128000, MaxOutput: 64000, Reasoning: true, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://api.individual.githubcopilot.com"},
 	// ----- google -----
 	{Provider: "google", ID: "gemini-3-flash-preview", DisplayName: "Gemini 3 Flash Preview", ContextWindow: 1048576, MaxOutput: 65536, Reasoning: true, PriceInput: 0.5, PriceOutput: 3, PriceCacheRead: 0.05, BaseURL: "https://generativelanguage.googleapis.com/v1beta"},
 	{Provider: "google", ID: "gemini-3-pro-preview", DisplayName: "Gemini 3 Pro Preview", ContextWindow: 1048576, MaxOutput: 65536, Reasoning: true, PriceInput: 2, PriceOutput: 12, PriceCacheRead: 0.2, BaseURL: "https://generativelanguage.googleapis.com/v1beta"},

A packages/provider/copilot_catalog_test.go +72 -0

@@ -0,0 +1,72 @@
	1	+	package provider
	2	+
	3	+	import (
	4	+	"context"
	5	+	"testing"
	6	+	)
	7	+
	8	+	func TestCopilotCatalogCoversPublicChatModels(t *testing.T) {
	9	+	groups := map[string][]string{
	10	+	APIAnthropicMessages: {"claude-fable-5", "claude-fable-5.1", "claude-haiku-4.5", "claude-opus-4.7", "claude-opus-4.8", "claude-opus-4.8-fast", "claude-opus-5", "claude-sonnet-5"},
	11	+	APIResponses: {"gpt-5-mini", "gpt-5.3-codex", "gpt-5.4", "gpt-5.4-mini", "gpt-5.5", "gpt-5.6-luna", "gpt-5.6-sol", "gpt-5.6-terra", "gpt-6-astra", "grok-4.5", "grok-4.6", "mai-code-1-flash-picker", "mai-code-1.1-flash"},
	12	+	APICompletions: {"gemini-3.5-flash", "gemini-3.6-flash", "gemini-3.7-flash", "gemini-3.8-flash", "kimi-k2.7-code", "kimi-k3"},
	13	+	}
	14	+	for api, ids := range groups {
	15	+	for _, id := range ids {
	16	+	t.Run(id, func(t *testing.T) {
	17	+	m, err := FindModel("github-copilot", id)
	18	+	if err != nil {
	19	+	t.Fatal(err)
	20	+	}
	21	+	if m.API != api {
	22	+	t.Fatalf("API = %q, want %q", m.API, api)
	23	+	}
	24	+	if m.ContextWindow <= 0 \|\| m.MaxOutput <= 0 \|\| m.MaxOutput >= m.ContextWindow {
	25	+	t.Fatalf("invalid context/output limits: %d/%d", m.ContextWindow, m.MaxOutput)
	26	+	}
	27	+	if m.BaseURL != copilotDefaultBaseURL \|\| !m.Reasoning {
	28	+	t.Fatal("invalid endpoint or reasoning metadata")
	29	+	}
	30	+	client := NewGithubCopilotClient("synthetic-token").(*copilotClient)
	31	+	captures := make(map[string]*routeCaptureClient)
	32	+	for _, key := range []string{APIAnthropicMessages, APIResponses, APICompletions} {
	33	+	capture := &routeCaptureClient{name: "github-copilot"}
	34	+	captures[key] = capture
	35	+	client.router.byAPI[key] = capture
	36	+	}
	37	+	stream, err := client.Stream(context.Background(), Request{Model: id})
	38	+	if err != nil {
	39	+	t.Fatal(err)
	40	+	}
	41	+	for range stream {
	42	+	}
	43	+	for key, capture := range captures {
	44	+	want := 0
	45	+	if key == api {
	46	+	want = 1
	47	+	}
	48	+	if len(capture.models) != want {
	49	+	t.Errorf("%s routed to %s incorrectly", id, key)
	50	+	}
	51	+	}
	52	+	})
	53	+	}
	54	+	}
	55	+	}
	56	+
	57	+	func TestCopilotCatalogExcludesRetiredAndUtilityModels(t *testing.T) {
	58	+	// https://docs.github.com/en/copilot/reference/ai-models/supported-models
	59	+	// Sonnet 4.6 is excluded from the general catalog despite its annual-plan exception.
	60	+	excluded := map[string]bool{
	61	+	"claude-opus-4.5": true, "claude-opus-4.6": true,
	62	+	"claude-sonnet-4.5": true, "claude-sonnet-4.6": true,
	63	+	"gemini-2.5-pro": true, "gemini-3-flash-preview": true, "gemini-3.1-pro-preview": true,
	64	+	"gpt-4.1": true, "gpt-4o": true, "gpt-5.2": true, "gpt-5.2-codex": true,
	65	+	"grok-code-fast-1": true, "gpt-5.4-nano": true,
	66	+	}
	67	+	for _, m := range Catalog {
	68	+	if m.Provider == "github-copilot" && excluded[m.ID] {
	69	+	t.Errorf("Copilot catalog still includes %q", m.ID)
	70	+	}
	71	+	}
	72	+	}

@@ -0,0 +1,72 @@
	1	+	package provider
	2	+
	3	+	import (
	4	+	"context"
	5	+	"testing"
	6	+	)
	7	+
	8	+	func TestCopilotCatalogCoversPublicChatModels(t *testing.T) {
	9	+	groups := map[string][]string{
	10	+	APIAnthropicMessages: {"claude-fable-5", "claude-fable-5.1", "claude-haiku-4.5", "claude-opus-4.7", "claude-opus-4.8", "claude-opus-4.8-fast", "claude-opus-5", "claude-sonnet-5"},
	11	+	APIResponses: {"gpt-5-mini", "gpt-5.3-codex", "gpt-5.4", "gpt-5.4-mini", "gpt-5.5", "gpt-5.6-luna", "gpt-5.6-sol", "gpt-5.6-terra", "gpt-6-astra", "grok-4.5", "grok-4.6", "mai-code-1-flash-picker", "mai-code-1.1-flash"},
	12	+	APICompletions: {"gemini-3.5-flash", "gemini-3.6-flash", "gemini-3.7-flash", "gemini-3.8-flash", "kimi-k2.7-code", "kimi-k3"},
	13	+	}
	14	+	for api, ids := range groups {
	15	+	for _, id := range ids {
	16	+	t.Run(id, func(t *testing.T) {
	17	+	m, err := FindModel("github-copilot", id)
	18	+	if err != nil {
	19	+	t.Fatal(err)
	20	+	}
	21	+	if m.API != api {
	22	+	t.Fatalf("API = %q, want %q", m.API, api)
	23	+	}
	24	+	if m.ContextWindow <= 0 \|\| m.MaxOutput <= 0 \|\| m.MaxOutput >= m.ContextWindow {
	25	+	t.Fatalf("invalid context/output limits: %d/%d", m.ContextWindow, m.MaxOutput)
	26	+	}
	27	+	if m.BaseURL != copilotDefaultBaseURL \|\| !m.Reasoning {
	28	+	t.Fatal("invalid endpoint or reasoning metadata")
	29	+	}
	30	+	client := NewGithubCopilotClient("synthetic-token").(*copilotClient)
	31	+	captures := make(map[string]*routeCaptureClient)
	32	+	for _, key := range []string{APIAnthropicMessages, APIResponses, APICompletions} {
	33	+	capture := &routeCaptureClient{name: "github-copilot"}
	34	+	captures[key] = capture
	35	+	client.router.byAPI[key] = capture
	36	+	}
	37	+	stream, err := client.Stream(context.Background(), Request{Model: id})
	38	+	if err != nil {
	39	+	t.Fatal(err)
	40	+	}
	41	+	for range stream {
	42	+	}
	43	+	for key, capture := range captures {
	44	+	want := 0
	45	+	if key == api {
	46	+	want = 1
	47	+	}
	48	+	if len(capture.models) != want {
	49	+	t.Errorf("%s routed to %s incorrectly", id, key)
	50	+	}
	51	+	}
	52	+	})
	53	+	}
	54	+	}
	55	+	}
	56	+
	57	+	func TestCopilotCatalogExcludesRetiredAndUtilityModels(t *testing.T) {
	58	+	// https://docs.github.com/en/copilot/reference/ai-models/supported-models
	59	+	// Sonnet 4.6 is excluded from the general catalog despite its annual-plan exception.
	60	+	excluded := map[string]bool{
	61	+	"claude-opus-4.5": true, "claude-opus-4.6": true,
	62	+	"claude-sonnet-4.5": true, "claude-sonnet-4.6": true,
	63	+	"gemini-2.5-pro": true, "gemini-3-flash-preview": true, "gemini-3.1-pro-preview": true,
	64	+	"gpt-4.1": true, "gpt-4o": true, "gpt-5.2": true, "gpt-5.2-codex": true,
	65	+	"grok-code-fast-1": true, "gpt-5.4-nano": true,
	66	+	}
	67	+	for _, m := range Catalog {
	68	+	if m.Provider == "github-copilot" && excluded[m.ID] {
	69	+	t.Errorf("Copilot catalog still includes %q", m.ID)
	70	+	}
	71	+	}
	72	+	}

A packages/provider/discover_copilot.go +83 -0

@@ -0,0 +1,83 @@
	1	+	package provider
	2	+
	3	+	import (
	4	+	"context"
	5	+	"encoding/json"
	6	+	"fmt"
	7	+	"io"
	8	+	"net/http"
	9	+	)
	10	+
	11	+	// DiscoverCopilotAvailableModels lists the account's selectable tool-capable
	12	+	// models. It does not enable model policies or persist account-specific data.
	13	+	func DiscoverCopilotAvailableModels(ctx context.Context, pat string) ([]string, error) {
	14	+	token, err := copilotCache.get(ctx, pat)
	15	+	if err != nil {
	16	+	return nil, err
	17	+	}
	18	+	baseURL := token.baseURL
	19	+	if baseURL == "" {
	20	+	baseURL = copilotDefaultBaseURL
	21	+	}
	22	+	client := &http.Client{Transport: &copilotRefreshTransport{inner: http.DefaultTransport, pat: pat}}
	23	+	req, err := http.NewRequestWithContext(ctx, http.MethodGet, baseURL+"/models", nil)
	24	+	if err != nil {
	25	+	return nil, err
	26	+	}
	27	+	req.Header.Set("Accept", "application/json")
	28	+	req.Header.Set("X-GitHub-Api-Version", "2026-06-01")
	29	+	resp, err := client.Do(req)
	30	+	if err != nil {
	31	+	return nil, fmt.Errorf("copilot model discovery: %w", err)
	32	+	}
	33	+	defer resp.Body.Close()
	34	+	if resp.StatusCode != http.StatusOK {
	35	+	return nil, fmt.Errorf("copilot model discovery: HTTP %d", resp.StatusCode)
	36	+	}
	37	+	body, err := io.ReadAll(io.LimitReader(resp.Body, 8<<20))
	38	+	if err != nil {
	39	+	return nil, fmt.Errorf("copilot model discovery: %w", err)
	40	+	}
	41	+	// The fallback is deliberately restricted to the individual endpoint.
	42	+	individual := baseURL == copilotDefaultBaseURL
	43	+	return parseCopilotAvailableModels(body, individual)
	44	+	}
	45	+
	46	+	func parseCopilotAvailableModels(body []byte, allowPolicyFallback bool) ([]string, error) {
	47	+	var catalog struct {
	48	+	Data []struct {
	49	+	ID string `json:"id"`
	50	+	Picker bool `json:"model_picker_enabled"`
	51	+	Policy struct {
	52	+	State string `json:"state"`
	53	+	} `json:"policy"`
	54	+	Capabilities struct {
	55	+	Supports struct {
	56	+	ToolCalls *bool `json:"tool_calls"`
	57	+	} `json:"supports"`
	58	+	} `json:"capabilities"`
	59	+	} `json:"data"`
	60	+	}
	61	+	if err := json.Unmarshal(body, &catalog); err != nil \|\| catalog.Data == nil {
	62	+	return nil, fmt.Errorf("copilot model discovery: invalid models response")
	63	+	}
	64	+	picker := make([]string, 0)
	65	+	enabled := make([]string, 0)
	66	+	seen := make(map[string]bool)
	67	+	for _, m := range catalog.Data {
	68	+	if m.ID == "" \|\| seen[m.ID] \|\| (m.Capabilities.Supports.ToolCalls != nil && !*m.Capabilities.Supports.ToolCalls) {
	69	+	continue
	70	+	}
	71	+	seen[m.ID] = true
	72	+	if m.Picker && m.Policy.State != "disabled" {
	73	+	picker = append(picker, m.ID)
	74	+	}
	75	+	if m.Policy.State == "enabled" {
	76	+	enabled = append(enabled, m.ID)
	77	+	}
	78	+	}
	79	+	if len(picker) == 0 && allowPolicyFallback {
	80	+	return enabled, nil
	81	+	}
	82	+	return picker, nil
	83	+	}

@@ -0,0 +1,83 @@
	1	+	package provider
	2	+
	3	+	import (
	4	+	"context"
	5	+	"encoding/json"
	6	+	"fmt"
	7	+	"io"
	8	+	"net/http"
	9	+	)
	10	+
	11	+	// DiscoverCopilotAvailableModels lists the account's selectable tool-capable
	12	+	// models. It does not enable model policies or persist account-specific data.
	13	+	func DiscoverCopilotAvailableModels(ctx context.Context, pat string) ([]string, error) {
	14	+	token, err := copilotCache.get(ctx, pat)
	15	+	if err != nil {
	16	+	return nil, err
	17	+	}
	18	+	baseURL := token.baseURL
	19	+	if baseURL == "" {
	20	+	baseURL = copilotDefaultBaseURL
	21	+	}
	22	+	client := &http.Client{Transport: &copilotRefreshTransport{inner: http.DefaultTransport, pat: pat}}
	23	+	req, err := http.NewRequestWithContext(ctx, http.MethodGet, baseURL+"/models", nil)
	24	+	if err != nil {
	25	+	return nil, err
	26	+	}
	27	+	req.Header.Set("Accept", "application/json")
	28	+	req.Header.Set("X-GitHub-Api-Version", "2026-06-01")
	29	+	resp, err := client.Do(req)
	30	+	if err != nil {
	31	+	return nil, fmt.Errorf("copilot model discovery: %w", err)
	32	+	}
	33	+	defer resp.Body.Close()
	34	+	if resp.StatusCode != http.StatusOK {
	35	+	return nil, fmt.Errorf("copilot model discovery: HTTP %d", resp.StatusCode)
	36	+	}
	37	+	body, err := io.ReadAll(io.LimitReader(resp.Body, 8<<20))
	38	+	if err != nil {
	39	+	return nil, fmt.Errorf("copilot model discovery: %w", err)
	40	+	}
	41	+	// The fallback is deliberately restricted to the individual endpoint.
	42	+	individual := baseURL == copilotDefaultBaseURL
	43	+	return parseCopilotAvailableModels(body, individual)
	44	+	}
	45	+
	46	+	func parseCopilotAvailableModels(body []byte, allowPolicyFallback bool) ([]string, error) {
	47	+	var catalog struct {
	48	+	Data []struct {
	49	+	ID string `json:"id"`
	50	+	Picker bool `json:"model_picker_enabled"`
	51	+	Policy struct {
	52	+	State string `json:"state"`
	53	+	} `json:"policy"`
	54	+	Capabilities struct {
	55	+	Supports struct {
	56	+	ToolCalls *bool `json:"tool_calls"`
	57	+	} `json:"supports"`
	58	+	} `json:"capabilities"`
	59	+	} `json:"data"`
	60	+	}
	61	+	if err := json.Unmarshal(body, &catalog); err != nil \|\| catalog.Data == nil {
	62	+	return nil, fmt.Errorf("copilot model discovery: invalid models response")
	63	+	}
	64	+	picker := make([]string, 0)
	65	+	enabled := make([]string, 0)
	66	+	seen := make(map[string]bool)
	67	+	for _, m := range catalog.Data {
	68	+	if m.ID == "" \|\| seen[m.ID] \|\| (m.Capabilities.Supports.ToolCalls != nil && !*m.Capabilities.Supports.ToolCalls) {
	69	+	continue
	70	+	}
	71	+	seen[m.ID] = true
	72	+	if m.Picker && m.Policy.State != "disabled" {
	73	+	picker = append(picker, m.ID)
	74	+	}
	75	+	if m.Policy.State == "enabled" {
	76	+	enabled = append(enabled, m.ID)
	77	+	}
	78	+	}
	79	+	if len(picker) == 0 && allowPolicyFallback {
	80	+	return enabled, nil
	81	+	}
	82	+	return picker, nil
	83	+	}

A packages/provider/discover_copilot_test.go +103 -0

@@ -0,0 +1,103 @@
	1	+	package provider
	2	+
	3	+	import (
	4	+	"context"
	5	+	"fmt"
	6	+	"net/http"
	7	+	"net/http/httptest"
	8	+	"reflect"
	9	+	"strings"
	10	+	"testing"
	11	+	)
	12	+
	13	+	func TestParseCopilotAvailableModels(t *testing.T) {
	14	+	for _, tc := range []struct {
	15	+	name, body string
	16	+	fallback bool
	17	+	want []string
	18	+	invalid bool
	19	+	}{
	20	+	{"picker", `{"data":[{"id":"yes","model_picker_enabled":true},{"id":"disabled","model_picker_enabled":true,"policy":{"state":"disabled"}},{"id":"hidden","policy":{"state":"enabled"}},{"id":"no-tools","model_picker_enabled":true,"capabilities":{"supports":{"tool_calls":false}}}]}`, true, []string{"yes"}, false},
	21	+	{"individual fallback", `{"data":[{"id":"enabled","policy":{"state":"enabled"}},{"id":"unconfigured","policy":{"state":"unconfigured"}},{"id":"disabled","policy":{"state":"disabled"}}]}`, true, []string{"enabled"}, false},
	22	+	{"enterprise no fallback", `{"data":[{"id":"enabled","policy":{"state":"enabled"}}]}`, false, []string{}, false},
	23	+	{"empty", `{"data":[]}`, true, []string{}, false},
	24	+	{"duplicate", `{"data":[{"id":"yes","model_picker_enabled":true},{"id":"yes","model_picker_enabled":true}]}`, false, []string{"yes"}, false},
	25	+	{"missing", `{}`, false, nil, true},
	26	+	{"null", `{"data":null}`, false, nil, true},
	27	+	{"invalid", `{"data":{}}`, false, nil, true},
	28	+	} {
	29	+	t.Run(tc.name, func(t *testing.T) {
	30	+	got, err := parseCopilotAvailableModels([]byte(tc.body), tc.fallback)
	31	+	if (err != nil) != tc.invalid {
	32	+	t.Fatalf("error = %v", err)
	33	+	}
	34	+	if !tc.invalid && !reflect.DeepEqual(got, tc.want) {
	35	+	t.Fatalf("got %v, want %v", got, tc.want)
	36	+	}
	37	+	})
	38	+	}
	39	+	}
	40	+
	41	+	func TestCopilotDiscoveryHTTP(t *testing.T) {
	42	+	for _, status := range []int{200, 403, 429} {
	43	+	t.Run(fmt.Sprint(status), func(t *testing.T) {
	44	+	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
	45	+	if r.Method != "GET" \|\| r.URL.Path != "/models" {
	46	+	t.Error("incorrect discovery request")
	47	+	}
	48	+	if r.Header.Get("Authorization") != "Bearer synthetic-session-token" \|\| r.Header.Get("Copilot-Integration-Id") != "vscode-chat" \|\| r.Header.Get("X-GitHub-Api-Version") != "2026-06-01" {
	49	+	t.Error("incorrect discovery headers")
	50	+	}
	51	+	w.WriteHeader(status)
	52	+	if status != 200 {
	53	+	fmt.Fprint(w, "sensitive-response-body")
	54	+	return
	55	+	}
	56	+	fmt.Fprint(w, `{"data":[{"id":"claude-sonnet-5","model_picker_enabled":true}]}`)
	57	+	}))
	58	+	defer srv.Close()
	59	+	pat := seedCopilotTestToken(t, srv.URL)
	60	+	got, err := DiscoverCopilotAvailableModels(context.Background(), pat)
	61	+	if status == 200 {
	62	+	if err != nil \|\| !reflect.DeepEqual(got, []string{"claude-sonnet-5"}) {
	63	+	t.Fatalf("got %v, %v", got, err)
	64	+	}
	65	+	} else if err == nil \|\| strings.Contains(err.Error(), "sensitive-response-body") {
	66	+	t.Fatalf("unexpected error %v", err)
	67	+	}
	68	+	ctx, cancel := context.WithCancel(context.Background())
	69	+	cancel()
	70	+	if _, err := DiscoverCopilotAvailableModels(ctx, pat); err == nil {
	71	+	t.Error("cancellation ignored")
	72	+	}
	73	+	})
	74	+	}
	75	+	}
	76	+
	77	+	func TestModelAvailabilityFiltersWithoutDeletingCatalog(t *testing.T) {
	78	+	t.Cleanup(func() { SetModelAvailability("github-copilot", nil) })
	79	+	before := len(ModelsForProvider("anthropic"))
	80	+	ids := []string{"claude-sonnet-5"}
	81	+	SetModelAvailability("github-copilot", ids)
	82	+	ids[0] = "mutated"
	83	+	models := ModelsForProvider("github-copilot")
	84	+	if len(models) != 1 \|\| models[0].ID != "claude-sonnet-5" {
	85	+	t.Fatalf("models = %v", models)
	86	+	}
	87	+	if len(ModelsForProvider("anthropic")) != before {
	88	+	t.Error("changed another provider")
	89	+	}
	90	+	SetModelAvailability("github-copilot", []string{})
	91	+	// Hiding a model must not change wire metadata for a session already using it.
	92	+	model, err := FindModel("github-copilot", "gpt-5.4")
	93	+	if err != nil \|\| model.API != APIResponses {
	94	+	t.Fatalf("availability refresh lost routing metadata: %+v, %v", model, err)
	95	+	}
	96	+	if len(ModelsForProvider("github-copilot")) != 0 {
	97	+	t.Error("empty availability did not hide models")
	98	+	}
	99	+	SetModelAvailability("github-copilot", nil)
	100	+	if len(ModelsForProvider("github-copilot")) != 27 {
	101	+	t.Error("clearing restriction did not restore catalog")
	102	+	}
	103	+	}

@@ -0,0 +1,103 @@
	1	+	package provider
	2	+
	3	+	import (
	4	+	"context"
	5	+	"fmt"
	6	+	"net/http"
	7	+	"net/http/httptest"
	8	+	"reflect"
	9	+	"strings"
	10	+	"testing"
	11	+	)
	12	+
	13	+	func TestParseCopilotAvailableModels(t *testing.T) {
	14	+	for _, tc := range []struct {
	15	+	name, body string
	16	+	fallback bool
	17	+	want []string
	18	+	invalid bool
	19	+	}{
	20	+	{"picker", `{"data":[{"id":"yes","model_picker_enabled":true},{"id":"disabled","model_picker_enabled":true,"policy":{"state":"disabled"}},{"id":"hidden","policy":{"state":"enabled"}},{"id":"no-tools","model_picker_enabled":true,"capabilities":{"supports":{"tool_calls":false}}}]}`, true, []string{"yes"}, false},
	21	+	{"individual fallback", `{"data":[{"id":"enabled","policy":{"state":"enabled"}},{"id":"unconfigured","policy":{"state":"unconfigured"}},{"id":"disabled","policy":{"state":"disabled"}}]}`, true, []string{"enabled"}, false},
	22	+	{"enterprise no fallback", `{"data":[{"id":"enabled","policy":{"state":"enabled"}}]}`, false, []string{}, false},
	23	+	{"empty", `{"data":[]}`, true, []string{}, false},
	24	+	{"duplicate", `{"data":[{"id":"yes","model_picker_enabled":true},{"id":"yes","model_picker_enabled":true}]}`, false, []string{"yes"}, false},
	25	+	{"missing", `{}`, false, nil, true},
	26	+	{"null", `{"data":null}`, false, nil, true},
	27	+	{"invalid", `{"data":{}}`, false, nil, true},
	28	+	} {
	29	+	t.Run(tc.name, func(t *testing.T) {
	30	+	got, err := parseCopilotAvailableModels([]byte(tc.body), tc.fallback)
	31	+	if (err != nil) != tc.invalid {
	32	+	t.Fatalf("error = %v", err)
	33	+	}
	34	+	if !tc.invalid && !reflect.DeepEqual(got, tc.want) {
	35	+	t.Fatalf("got %v, want %v", got, tc.want)
	36	+	}
	37	+	})
	38	+	}
	39	+	}
	40	+
	41	+	func TestCopilotDiscoveryHTTP(t *testing.T) {
	42	+	for _, status := range []int{200, 403, 429} {
	43	+	t.Run(fmt.Sprint(status), func(t *testing.T) {
	44	+	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
	45	+	if r.Method != "GET" \|\| r.URL.Path != "/models" {
	46	+	t.Error("incorrect discovery request")
	47	+	}
	48	+	if r.Header.Get("Authorization") != "Bearer synthetic-session-token" \|\| r.Header.Get("Copilot-Integration-Id") != "vscode-chat" \|\| r.Header.Get("X-GitHub-Api-Version") != "2026-06-01" {
	49	+	t.Error("incorrect discovery headers")
	50	+	}
	51	+	w.WriteHeader(status)
	52	+	if status != 200 {
	53	+	fmt.Fprint(w, "sensitive-response-body")
	54	+	return
	55	+	}
	56	+	fmt.Fprint(w, `{"data":[{"id":"claude-sonnet-5","model_picker_enabled":true}]}`)
	57	+	}))
	58	+	defer srv.Close()
	59	+	pat := seedCopilotTestToken(t, srv.URL)
	60	+	got, err := DiscoverCopilotAvailableModels(context.Background(), pat)
	61	+	if status == 200 {
	62	+	if err != nil \|\| !reflect.DeepEqual(got, []string{"claude-sonnet-5"}) {
	63	+	t.Fatalf("got %v, %v", got, err)
	64	+	}
	65	+	} else if err == nil \|\| strings.Contains(err.Error(), "sensitive-response-body") {
	66	+	t.Fatalf("unexpected error %v", err)
	67	+	}
	68	+	ctx, cancel := context.WithCancel(context.Background())
	69	+	cancel()
	70	+	if _, err := DiscoverCopilotAvailableModels(ctx, pat); err == nil {
	71	+	t.Error("cancellation ignored")
	72	+	}
	73	+	})
	74	+	}
	75	+	}
	76	+
	77	+	func TestModelAvailabilityFiltersWithoutDeletingCatalog(t *testing.T) {
	78	+	t.Cleanup(func() { SetModelAvailability("github-copilot", nil) })
	79	+	before := len(ModelsForProvider("anthropic"))
	80	+	ids := []string{"claude-sonnet-5"}
	81	+	SetModelAvailability("github-copilot", ids)
	82	+	ids[0] = "mutated"
	83	+	models := ModelsForProvider("github-copilot")
	84	+	if len(models) != 1 \|\| models[0].ID != "claude-sonnet-5" {
	85	+	t.Fatalf("models = %v", models)
	86	+	}
	87	+	if len(ModelsForProvider("anthropic")) != before {
	88	+	t.Error("changed another provider")
	89	+	}
	90	+	SetModelAvailability("github-copilot", []string{})
	91	+	// Hiding a model must not change wire metadata for a session already using it.
	92	+	model, err := FindModel("github-copilot", "gpt-5.4")
	93	+	if err != nil \|\| model.API != APIResponses {
	94	+	t.Fatalf("availability refresh lost routing metadata: %+v, %v", model, err)
	95	+	}
	96	+	if len(ModelsForProvider("github-copilot")) != 0 {
	97	+	t.Error("empty availability did not hide models")
	98	+	}
	99	+	SetModelAvailability("github-copilot", nil)
	100	+	if len(ModelsForProvider("github-copilot")) != 27 {
	101	+	t.Error("clearing restriction did not restore catalog")
	102	+	}
	103	+	}

M packages/provider/extra_providers.go +2 -1


@@ -337,7 +337,8 @@ func NewCloudflareAIGateway(apiKey, baseURL string) Client {
337	337		// access enabled; zot trades it for a short-lived Copilot token on
338	338		// every inference request (cached in memory until ~5min before expiry).
339	339		//
340		-	// Wire format: OpenAI Chat Completions. Copilot-specific headers
	340	+	// Wire formats: Anthropic Messages, OpenAI Responses, and Chat Completions,
	341	+	// selected by model family. Copilot-specific headers
341	342		// (X-Initiator, Openai-Intent, Editor-Version, Editor-Plugin-Version,
342	343		// Copilot-Integration-Id, User-Agent) are added by the refresh
343	344		// transport. The model id passes through unchanged.

 // access enabled; zot trades it for a short-lived Copilot token on
 // every inference request (cached in memory until ~5min before expiry).
 //
-// Wire format: OpenAI Chat Completions. Copilot-specific headers
+// Wire formats: Anthropic Messages, OpenAI Responses, and Chat Completions,
+// selected by model family. Copilot-specific headers
 // (X-Initiator, Openai-Intent, Editor-Version, Editor-Plugin-Version,
 // Copilot-Integration-Id, User-Agent) are added by the refresh
 // transport. The model id passes through unchanged.

M packages/provider/github_copilot.go +26 -16


@@ -14,12 +14,11 @@ package provider
14	14		// User-Agent). The response carries `{ "token": "...", "expires_at": <unix> }`.
15	15		// 3. The short-lived token's value embeds a `proxy-ep=<host>` field that
16	16		// tells us the real API host (individual users: api.individual.githubcopilot.com).
17		-	// 4. Inference requests go to `<host>/chat/completions` with the
	17	+	// 4. Inference requests use Messages, Responses, or Chat Completions with the
18	18		// short-lived token in `Authorization: Bearer` plus extras:
19	19		// - X-Initiator: user\|agent
20	20		// - Openai-Intent: conversation-edits
21		-	// - Copilot-Vision-Request: true (when images present; not wired
22		-	// here because the zot openai client currently sends images inline)
	21	+	// - Copilot-Vision-Request: true (when images are present)
23	22		//
24	23		// Token caching: short-lived tokens last ~30min. We cache one per PAT in
25	24		// memory for the process lifetime and refresh on demand. No disk cache.

@@ -162,12 +161,21 @@ func (t copilotRefreshTransport) RoundTrip(req http.Request) (*http.Response,
162	161		}
163	162		clone := req.Clone(req.Context())
164	163		clone.Header.Set("Authorization", "Bearer "+tok.value)
	164	+	// Messages uses Copilot Bearer auth, never an Anthropic API key.
	165	+	clone.Header.Del("x-api-key")
165	166		// Identity headers also required on inference requests.
166	167		for k, v := range copilotIdentityHeaders {
167	168		clone.Header.Set(k, v)
168	169		}
169		-	clone.Header.Set("X-Initiator", "agent")
170		-	clone.Header.Set("Openai-Intent", "conversation-edits")
	170	+	if metadata, ok := req.Context().Value(copilotRequestKey{}).(copilotRequestMetadata); ok {
	171	+	clone.Header.Set("X-Initiator", metadata.initiator)
	172	+	if metadata.vision {
	173	+	clone.Header.Set("Copilot-Vision-Request", "true")
	174	+	} else {
	175	+	clone.Header.Del("Copilot-Vision-Request")
	176	+	}
	177	+	clone.Header.Set("Openai-Intent", "conversation-edits")
	178	+	}
171	179		// If the request URL host doesn't match the token's proxy-ep, rewrite
172	180		// it. The openaiClient pinned a static host at construction time, but
173	181		// the canonical host comes from the token.

@@ -199,14 +207,8 @@ func (t copilotResponsesStripTransport) RoundTrip(req http.Request) (*http.Res
199	207
200	208		const copilotDefaultBaseURL = "https://api.individual.githubcopilot.com"
201	209
202		-	// NewGithubCopilotClient returns a Copilot-pinned OpenAI-compat client.
203		-	// The pat must be a GitHub Personal Access Token with Copilot access.
204		-	//
205		-	// Copilot exposes two wire protocols: most models use Chat Completions
206		-	// (/chat/completions), while newer GPT models use the Responses API
207		-	// (/responses), including GPT-5.6 (sol/terra/luna) and GPT-6 Astra.
208		-	// A model router dispatches each request to the matching wire client
209		-	// based on the model's catalog API tag.
	210	+	// NewGithubCopilotClient uses Messages for Claude, Responses for GPT/Grok/MAI,
	211	+	// and Chat Completions for Gemini/Kimi. The PAT is exchanged for Copilot tokens.
210	212		func NewGithubCopilotClient(pat string) Client {
211	213		refresh := &copilotRefreshTransport{inner: http.DefaultTransport, pat: pat}
212	214		completionsHTTP := &http.Client{Transport: refresh, Timeout: 0}

@@ -233,7 +235,15 @@ func NewGithubCopilotClient(pat string) Client {
233	235		http: responsesHTTP,
234	236		}
235	237
236		-	return NewModelRouter("github-copilot", completions, map[string]Client{
237		-	APIResponses: &renamedClient{inner: responses, name: "github-copilot"},
238		-	})
	238	+	messages := &anthropicClient{
	239	+	name: "github-copilot",
	240	+	baseURL: copilotDefaultBaseURL,
	241	+	// Do not set oauthTok: that would inject Claude Code identity and rename tools.
	242	+	http: &http.Client{Transport: refresh},
	243	+	}
	244	+	return &copilotClient{router: NewModelRouter("github-copilot", completions, map[string]Client{
	245	+	APICompletions: completions,
	246	+	APIResponses: responses,
	247	+	APIAnthropicMessages: messages,
	248	+	}).(*modelRouter)}
239	249		}

 //     User-Agent). The response carries `{ "token": "...", "expires_at": <unix> }`.
 //  3. The short-lived token's value embeds a `proxy-ep=<host>` field that
 //     tells us the real API host (individual users: api.individual.githubcopilot.com).
-//  4. Inference requests go to `<host>/chat/completions` with the
+//  4. Inference requests use Messages, Responses, or Chat Completions with the
 //     short-lived token in `Authorization: Bearer` plus extras:
 //       - X-Initiator: user|agent
 //       - Openai-Intent: conversation-edits
-//       - Copilot-Vision-Request: true (when images present; not wired
+//       - Copilot-Vision-Request: true (when images are present)
-//         here because the zot openai client currently sends images inline)
 //
 // Token caching: short-lived tokens last ~30min. We cache one per PAT in
 // memory for the process lifetime and refresh on demand. No disk cache.
 	}
 	clone := req.Clone(req.Context())
 	clone.Header.Set("Authorization", "Bearer "+tok.value)
+	// Messages uses Copilot Bearer auth, never an Anthropic API key.
+	clone.Header.Del("x-api-key")
 	// Identity headers also required on inference requests.
 	for k, v := range copilotIdentityHeaders {
 		clone.Header.Set(k, v)
 	}
-	clone.Header.Set("X-Initiator", "agent")
+	if metadata, ok := req.Context().Value(copilotRequestKey{}).(copilotRequestMetadata); ok {
-	clone.Header.Set("Openai-Intent", "conversation-edits")
+		clone.Header.Set("X-Initiator", metadata.initiator)
+		if metadata.vision {
+			clone.Header.Set("Copilot-Vision-Request", "true")
+		} else {
+			clone.Header.Del("Copilot-Vision-Request")
+		}
+		clone.Header.Set("Openai-Intent", "conversation-edits")
+	}
 	// If the request URL host doesn't match the token's proxy-ep, rewrite
 	// it. The openaiClient pinned a static host at construction time, but
 	// the canonical host comes from the token.
 const copilotDefaultBaseURL = "https://api.individual.githubcopilot.com"
-// NewGithubCopilotClient returns a Copilot-pinned OpenAI-compat client.
+// NewGithubCopilotClient uses Messages for Claude, Responses for GPT/Grok/MAI,
-// The pat must be a GitHub Personal Access Token with Copilot access.
+// and Chat Completions for Gemini/Kimi. The PAT is exchanged for Copilot tokens.
-//
-// Copilot exposes two wire protocols: most models use Chat Completions
-// (/chat/completions), while newer GPT models use the Responses API
-// (/responses), including GPT-5.6 (sol/terra/luna) and GPT-6 Astra.
-// A model router dispatches each request to the matching wire client
-// based on the model's catalog API tag.
 func NewGithubCopilotClient(pat string) Client {
 	refresh := &copilotRefreshTransport{inner: http.DefaultTransport, pat: pat}
 	completionsHTTP := &http.Client{Transport: refresh, Timeout: 0}
 		http:              responsesHTTP,
 	}
-	return NewModelRouter("github-copilot", completions, map[string]Client{
+	messages := &anthropicClient{
-		APIResponses: &renamedClient{inner: responses, name: "github-copilot"},
+		name:    "github-copilot",
-	})
+		baseURL: copilotDefaultBaseURL,
+		// Do not set oauthTok: that would inject Claude Code identity and rename tools.
+		http: &http.Client{Transport: refresh},
+	}
+	return &copilotClient{router: NewModelRouter("github-copilot", completions, map[string]Client{
+		APICompletions:       completions,
+		APIResponses:         responses,
+		APIAnthropicMessages: messages,
+	}).(*modelRouter)}
 }

A packages/provider/github_copilot_request.go +83 -0

@@ -0,0 +1,83 @@
	1	+	package provider
	2	+
	3	+	import (
	4	+	"context"
	5	+	"strings"
	6	+	)
	7	+
	8	+	type copilotRequestKey struct{}
	9	+	type copilotRequestMetadata struct {
	10	+	initiator string
	11	+	vision bool
	12	+	}
	13	+
	14	+	type copilotClient struct {
	15	+	router *modelRouter
	16	+	}
	17	+
	18	+	func (c *copilotClient) Name() string { return "github-copilot" }
	19	+
	20	+	func (c *copilotClient) Stream(ctx context.Context, req Request) (<-chan Event, error) {
	21	+	metadata := copilotRequestMetadata{initiator: "user"}
	22	+	if len(req.Messages) > 0 && req.Messages[len(req.Messages)-1].Role != RoleUser {
	23	+	metadata.initiator = "agent"
	24	+	}
	25	+	for _, message := range req.Messages {
	26	+	if message.Role == RoleUser \|\| message.Role == RoleTool {
	27	+	metadata.vision = metadata.vision \|\| copilotHasImages(message.Content)
	28	+	}
	29	+	}
	30	+	// These endpoints perform their own thinking; they reject reasoning_effort.
	31	+	api := copilotModelAPI(req.Model)
	32	+	if model, err := FindModel(c.Name(), req.Model); err == nil && model.API != "" {
	33	+	api = model.API
	34	+	}
	35	+	if api == APICompletions {
	36	+	req.Reasoning = ""
	37	+	}
	38	+	ctx = context.WithValue(ctx, copilotRequestKey{}, metadata)
	39	+	// Use family routing for custom models that have not supplied an API tag.
	40	+	if model, err := FindModel(c.Name(), req.Model); err != nil \|\| model.API == "" {
	41	+	return c.router.byAPI[api].Stream(ctx, req)
	42	+	}
	43	+	return c.router.Stream(ctx, req)
	44	+	}
	45	+
	46	+	func copilotHasImages(blocks []Content) bool {
	47	+	for _, block := range blocks {
	48	+	switch b := block.(type) {
	49	+	case ImageBlock:
	50	+	return true
	51	+	case ToolResultBlock:
	52	+	if copilotHasImages(b.Content) {
	53	+	return true
	54	+	}
	55	+	}
	56	+	}
	57	+	return false
	58	+	}
	59	+
	60	+	func copilotModelAPI(id string) string {
	61	+	switch {
	62	+	case strings.HasPrefix(id, "claude-"):
	63	+	return APIAnthropicMessages
	64	+	case strings.HasPrefix(id, "gpt-"), strings.HasPrefix(id, "grok-"), strings.HasPrefix(id, "mai-"), strings.HasPrefix(id, "oswe"):
	65	+	return APIResponses
	66	+	default:
	67	+	return APICompletions
	68	+	}
	69	+	}
	70	+
	71	+	func configureCopilotModel(m *Model) {
	72	+	m.API = copilotModelAPI(m.ID)
	73	+	if m.API == APIAnthropicMessages && !strings.HasPrefix(m.ID, "claude-haiku-") {
	74	+	m.AdaptiveThinking = true
	75	+	}
	76	+	if m.API == APICompletions {
	77	+	m.ReasoningLevelMap = map[string]string{"minimum": "", "low": "", "medium": "", "high": "", "xhigh": "", "max": ""}
	78	+	}
	79	+	// Responses defaults expose xhigh, but these models only support low/medium/high.
	80	+	if m.ID == "gpt-5-mini" \|\| m.ID == "grok-4.5" \|\| strings.HasPrefix(m.ID, "mai-") {
	81	+	m.ReasoningLevelMap = map[string]string{"xhigh": "", "max": ""}
	82	+	}
	83	+	}

@@ -0,0 +1,83 @@
	1	+	package provider
	2	+
	3	+	import (
	4	+	"context"
	5	+	"strings"
	6	+	)
	7	+
	8	+	type copilotRequestKey struct{}
	9	+	type copilotRequestMetadata struct {
	10	+	initiator string
	11	+	vision bool
	12	+	}
	13	+
	14	+	type copilotClient struct {
	15	+	router *modelRouter
	16	+	}
	17	+
	18	+	func (c *copilotClient) Name() string { return "github-copilot" }
	19	+
	20	+	func (c *copilotClient) Stream(ctx context.Context, req Request) (<-chan Event, error) {
	21	+	metadata := copilotRequestMetadata{initiator: "user"}
	22	+	if len(req.Messages) > 0 && req.Messages[len(req.Messages)-1].Role != RoleUser {
	23	+	metadata.initiator = "agent"
	24	+	}
	25	+	for _, message := range req.Messages {
	26	+	if message.Role == RoleUser \|\| message.Role == RoleTool {
	27	+	metadata.vision = metadata.vision \|\| copilotHasImages(message.Content)
	28	+	}
	29	+	}
	30	+	// These endpoints perform their own thinking; they reject reasoning_effort.
	31	+	api := copilotModelAPI(req.Model)
	32	+	if model, err := FindModel(c.Name(), req.Model); err == nil && model.API != "" {
	33	+	api = model.API
	34	+	}
	35	+	if api == APICompletions {
	36	+	req.Reasoning = ""
	37	+	}
	38	+	ctx = context.WithValue(ctx, copilotRequestKey{}, metadata)
	39	+	// Use family routing for custom models that have not supplied an API tag.
	40	+	if model, err := FindModel(c.Name(), req.Model); err != nil \|\| model.API == "" {
	41	+	return c.router.byAPI[api].Stream(ctx, req)
	42	+	}
	43	+	return c.router.Stream(ctx, req)
	44	+	}
	45	+
	46	+	func copilotHasImages(blocks []Content) bool {
	47	+	for _, block := range blocks {
	48	+	switch b := block.(type) {
	49	+	case ImageBlock:
	50	+	return true
	51	+	case ToolResultBlock:
	52	+	if copilotHasImages(b.Content) {
	53	+	return true
	54	+	}
	55	+	}
	56	+	}
	57	+	return false
	58	+	}
	59	+
	60	+	func copilotModelAPI(id string) string {
	61	+	switch {
	62	+	case strings.HasPrefix(id, "claude-"):
	63	+	return APIAnthropicMessages
	64	+	case strings.HasPrefix(id, "gpt-"), strings.HasPrefix(id, "grok-"), strings.HasPrefix(id, "mai-"), strings.HasPrefix(id, "oswe"):
	65	+	return APIResponses
	66	+	default:
	67	+	return APICompletions
	68	+	}
	69	+	}
	70	+
	71	+	func configureCopilotModel(m *Model) {
	72	+	m.API = copilotModelAPI(m.ID)
	73	+	if m.API == APIAnthropicMessages && !strings.HasPrefix(m.ID, "claude-haiku-") {
	74	+	m.AdaptiveThinking = true
	75	+	}
	76	+	if m.API == APICompletions {
	77	+	m.ReasoningLevelMap = map[string]string{"minimum": "", "low": "", "medium": "", "high": "", "xhigh": "", "max": ""}
	78	+	}
	79	+	// Responses defaults expose xhigh, but these models only support low/medium/high.
	80	+	if m.ID == "gpt-5-mini" \|\| m.ID == "grok-4.5" \|\| strings.HasPrefix(m.ID, "mai-") {
	81	+	m.ReasoningLevelMap = map[string]string{"xhigh": "", "max": ""}
	82	+	}
	83	+	}

M packages/provider/github_copilot_test.go +8 -7


@@ -31,6 +31,11 @@ func TestCopilotHostRewrite(t *testing.T) {
31	31		baseURL: "https://api.enterprise.copilot.proxy",
32	32		}
33	33		copilotCache.mu.Unlock()
	34	+	t.Cleanup(func() {
	35	+	copilotCache.mu.Lock()
	36	+	delete(copilotCache.tokens, pat)
	37	+	copilotCache.mu.Unlock()
	38	+	})
34	39
35	40		mockRT := &mockRoundTripper{}
36	41		transport := &copilotRefreshTransport{

@@ -70,15 +75,11 @@ func TestCopilotHostRewrite(t *testing.T) {
70	75		}
71	76
72	77		func TestCopilotRoutesModernGPTToResponses(t *testing.T) {
73		-	router, ok := NewGithubCopilotClient("pat").(*modelRouter)
	78	+	client, ok := NewGithubCopilotClient("pat").(*copilotClient)
74	79		if !ok {
75		-	t.Fatal("NewGithubCopilotClient did not return a modelRouter")
	80	+	t.Fatal("NewGithubCopilotClient did not return a copilotClient")
76	81		}
77		-	responses, ok := router.byAPI[APIResponses].(*renamedClient)
78		-	if !ok {
79		-	t.Fatal("github-copilot router has no Responses client")
80		-	}
81		-	codex, ok := responses.inner.(*codexClient)
	82	+	codex, ok := client.router.byAPI[APIResponses].(*codexClient)
82	83		if !ok {
83	84		t.Fatal("github-copilot Responses client is not a codexClient")
84	85		}

 		baseURL:   "https://api.enterprise.copilot.proxy",
 	}
 	copilotCache.mu.Unlock()
+	t.Cleanup(func() {
+		copilotCache.mu.Lock()
+		delete(copilotCache.tokens, pat)
+		copilotCache.mu.Unlock()
+	})
 	mockRT := &mockRoundTripper{}
 	transport := &copilotRefreshTransport{
 }
 func TestCopilotRoutesModernGPTToResponses(t *testing.T) {
-	router, ok := NewGithubCopilotClient("pat").(*modelRouter)
+	client, ok := NewGithubCopilotClient("pat").(*copilotClient)
 	if !ok {
-		t.Fatal("NewGithubCopilotClient did not return a modelRouter")
+		t.Fatal("NewGithubCopilotClient did not return a copilotClient")
 	}
-	responses, ok := router.byAPI[APIResponses].(*renamedClient)
+	codex, ok := client.router.byAPI[APIResponses].(*codexClient)
-	if !ok {
-		t.Fatal("github-copilot router has no Responses client")
-	}
-	codex, ok := responses.inner.(*codexClient)
 	if !ok {
 		t.Fatal("github-copilot Responses client is not a codexClient")
 	}

A packages/provider/github_copilot_wire_test.go +174 -0

@@ -0,0 +1,174 @@
	1	+	package provider
	2	+
	3	+	import (
	4	+	"context"
	5	+	"encoding/json"
	6	+	"fmt"
	7	+	"net/http"
	8	+	"net/http/httptest"
	9	+	"strings"
	10	+	"testing"
	11	+	"time"
	12	+	)
	13	+
	14	+	func seedCopilotTestToken(t *testing.T, baseURL string) string {
	15	+	t.Helper()
	16	+	pat := "synthetic-" + t.Name()
	17	+	copilotCache.mu.Lock()
	18	+	copilotCache.tokens[pat] = copilotToken{value: "synthetic-session-token", baseURL: baseURL, expiresAt: time.Now().Add(time.Hour)}
	19	+	copilotCache.mu.Unlock()
	20	+	t.Cleanup(func() {
	21	+	copilotCache.mu.Lock()
	22	+	delete(copilotCache.tokens, pat)
	23	+	copilotCache.mu.Unlock()
	24	+	})
	25	+	return pat
	26	+	}
	27	+
	28	+	func TestCopilotWireProtocols(t *testing.T) {
	29	+	for _, tc := range []struct{ model, path string }{
	30	+	{"claude-sonnet-5", "/v1/messages"},
	31	+	{"claude-haiku-4.5", "/v1/messages"},
	32	+	{"gpt-5.4", "/responses"},
	33	+	{"grok-4.6", "/responses"},
	34	+	{"mai-code-1.1-flash", "/responses"},
	35	+	{"gemini-3.8-flash", "/chat/completions"},
	36	+	{"kimi-k3", "/chat/completions"},
	37	+	} {
	38	+	t.Run(tc.model, func(t *testing.T) {
	39	+	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
	40	+	if r.URL.Path != tc.path {
	41	+	t.Errorf("path = %q, want %q", r.URL.Path, tc.path)
	42	+	}
	43	+	if r.Header.Get("Authorization") != "Bearer synthetic-session-token" {
	44	+	t.Error("missing session Bearer auth")
	45	+	}
	46	+	if r.Header.Get("Copilot-Integration-Id") != "vscode-chat" \|\| r.Header.Get("X-Initiator") != "user" \|\| r.Header.Get("Openai-Intent") != "conversation-edits" {
	47	+	t.Error("missing Copilot identity headers")
	48	+	}
	49	+	for _, header := range []string{"x-api-key", "anthropic-beta", "x-app", "chatgpt-account-id", "openai-beta", "originator"} {
	50	+	if r.Header.Get(header) != "" {
	51	+	t.Errorf("unexpected identity header %s", header)
	52	+	}
	53	+	}
	54	+	var body map[string]json.RawMessage
	55	+	if err := json.NewDecoder(r.Body).Decode(&body); err != nil {
	56	+	t.Error(err)
	57	+	w.WriteHeader(400)
	58	+	return
	59	+	}
	60	+	if string(body["model"]) != fmt.Sprintf("%q", tc.model) {
	61	+	t.Error("model ID changed on wire")
	62	+	}
	63	+	w.Header().Set("Content-Type", "text/event-stream")
	64	+	switch tc.path {
	65	+	case "/v1/messages":
	66	+	if len(body["messages"]) == 0 \|\| len(body["max_tokens"]) == 0 {
	67	+	t.Error("missing Messages fields")
	68	+	}
	69	+	if strings.Contains(string(body["system"]), "Claude Code") {
	70	+	t.Error("injected Claude Code system identity")
	71	+	}
	72	+	if tc.model == "claude-sonnet-5" {
	73	+	if string(body["thinking"]) != `{"type":"adaptive"}` {
	74	+	t.Errorf("thinking = %s", body["thinking"])
	75	+	}
	76	+	if len(body["output_config"]) == 0 \|\| len(body["temperature"]) != 0 {
	77	+	t.Error("invalid adaptive thinking payload")
	78	+	}
	79	+	}
	80	+	fmt.Fprint(w, "event: message_start\ndata: {\"message\":{\"id\":\"test\",\"usage\":{\"input_tokens\":1}}}\n\nevent: message_delta\ndata: {\"delta\":{\"stop_reason\":\"end_turn\"},\"usage\":{\"output_tokens\":1}}\n\nevent: message_stop\ndata: {}\n\n")
	81	+	case "/responses":
	82	+	if len(body["input"]) == 0 \|\| len(body["reasoning"]) == 0 {
	83	+	t.Error("missing Responses fields")
	84	+	}
	85	+	fmt.Fprint(w, "data: {\"type\":\"response.completed\",\"response\":{\"status\":\"completed\",\"usage\":{\"input_tokens\":1,\"output_tokens\":1}}}\n\n")
	86	+	default:
	87	+	if len(body["reasoning_effort"]) != 0 {
	88	+	t.Error("unsupported reasoning_effort sent")
	89	+	}
	90	+	if len(body["messages"]) == 0 {
	91	+	t.Error("missing chat messages")
	92	+	}
	93	+	fmt.Fprint(w, "data: {\"choices\":[{\"index\":0,\"delta\":{\"content\":\"ok\"},\"finish_reason\":\"stop\"}]}\n\ndata: [DONE]\n\n")
	94	+	}
	95	+	}))
	96	+	defer srv.Close()
	97	+	client := NewGithubCopilotClient(seedCopilotTestToken(t, srv.URL))
	98	+	stream, err := client.Stream(context.Background(), Request{Model: tc.model, Reasoning: "high", System: "Test system", Messages: []Message{{Role: RoleUser, Content: []Content{TextBlock{Text: "Hello"}}}}})
	99	+	if err != nil {
	100	+	t.Fatal(err)
	101	+	}
	102	+	for event := range stream {
	103	+	if e, ok := event.(EventDone); ok && e.Err != nil {
	104	+	t.Fatalf("stream error: %v", e.Err)
	105	+	}
	106	+	if e, ok := event.(EventStart); ok && e.Provider != "github-copilot" {
	107	+	t.Errorf("event provider = %s", e.Provider)
	108	+	}
	109	+	}
	110	+	})
	111	+	}
	112	+	}
	113	+
	114	+	type copilotMetadataCapture struct {
	115	+	metadata copilotRequestMetadata
	116	+	reasoning string
	117	+	}
	118	+
	119	+	func (c *copilotMetadataCapture) Name() string { return "github-copilot" }
	120	+	func (c *copilotMetadataCapture) Stream(ctx context.Context, req Request) (<-chan Event, error) {
	121	+	c.metadata = ctx.Value(copilotRequestKey{}).(copilotRequestMetadata)
	122	+	c.reasoning = req.Reasoning
	123	+	ch := make(chan Event)
	124	+	close(ch)
	125	+	return ch, nil
	126	+	}
	127	+
	128	+	func TestCopilotDynamicRequestMetadata(t *testing.T) {
	129	+	client := NewGithubCopilotClient("unused").(*copilotClient)
	130	+	capture := &copilotMetadataCapture{}
	131	+	client.router.byAPI[APICompletions] = capture
	132	+	for _, tc := range []struct {
	133	+	messages []Message
	134	+	initiator string
	135	+	vision bool
	136	+	}{
	137	+	{nil, "user", false},
	138	+	{[]Message{{Role: RoleUser, Content: []Content{ImageBlock{}}}}, "user", true},
	139	+	{[]Message{{Role: RoleTool, Content: []Content{ToolResultBlock{Content: []Content{ImageBlock{}}}}}}, "agent", true},
	140	+	{[]Message{{Role: RoleAssistant}}, "agent", false},
	141	+	{[]Message{{Role: RoleUser}}, "user", false},
	142	+	} {
	143	+	_, err := client.Stream(context.Background(), Request{Model: "kimi-k3", Reasoning: "high", Messages: tc.messages})
	144	+	if err != nil {
	145	+	t.Fatal(err)
	146	+	}
	147	+	if capture.metadata.initiator != tc.initiator \|\| capture.metadata.vision != tc.vision \|\| capture.reasoning != "" {
	148	+	t.Errorf("unexpected metadata: %+v, reasoning %q", capture.metadata, capture.reasoning)
	149	+	}
	150	+	}
	151	+	// The transport must propagate vision metadata and remove it on the next text-only request.
	152	+	mock := &mockRoundTripper{}
	153	+	transport := &copilotRefreshTransport{inner: mock, pat: seedCopilotTestToken(t, copilotDefaultBaseURL)}
	154	+	for _, vision := range []bool{true, false} {
	155	+	ctx := context.WithValue(context.Background(), copilotRequestKey{}, copilotRequestMetadata{initiator: "agent", vision: vision})
	156	+	req, _ := http.NewRequestWithContext(ctx, "POST", copilotDefaultBaseURL+"/responses", nil)
	157	+	req.Header.Set("Copilot-Vision-Request", "stale")
	158	+	resp, err := transport.RoundTrip(req)
	159	+	if err != nil {
	160	+	t.Fatal(err)
	161	+	}
	162	+	resp.Body.Close()
	163	+	want := ""
	164	+	if vision {
	165	+	want = "true"
	166	+	}
	167	+	if mock.lastReq.Header.Get("Copilot-Vision-Request") != want \|\| mock.lastReq.Header.Get("X-Initiator") != "agent" {
	168	+	t.Error("incorrect dynamic transport headers")
	169	+	}
	170	+	if req.Header.Get("Copilot-Vision-Request") != "stale" {
	171	+	t.Error("transport mutated caller headers")
	172	+	}
	173	+	}
	174	+	}

@@ -0,0 +1,174 @@
	1	+	package provider
	2	+
	3	+	import (
	4	+	"context"
	5	+	"encoding/json"
	6	+	"fmt"
	7	+	"net/http"
	8	+	"net/http/httptest"
	9	+	"strings"
	10	+	"testing"
	11	+	"time"
	12	+	)
	13	+
	14	+	func seedCopilotTestToken(t *testing.T, baseURL string) string {
	15	+	t.Helper()
	16	+	pat := "synthetic-" + t.Name()
	17	+	copilotCache.mu.Lock()
	18	+	copilotCache.tokens[pat] = copilotToken{value: "synthetic-session-token", baseURL: baseURL, expiresAt: time.Now().Add(time.Hour)}
	19	+	copilotCache.mu.Unlock()
	20	+	t.Cleanup(func() {
	21	+	copilotCache.mu.Lock()
	22	+	delete(copilotCache.tokens, pat)
	23	+	copilotCache.mu.Unlock()
	24	+	})
	25	+	return pat
	26	+	}
	27	+
	28	+	func TestCopilotWireProtocols(t *testing.T) {
	29	+	for _, tc := range []struct{ model, path string }{
	30	+	{"claude-sonnet-5", "/v1/messages"},
	31	+	{"claude-haiku-4.5", "/v1/messages"},
	32	+	{"gpt-5.4", "/responses"},
	33	+	{"grok-4.6", "/responses"},
	34	+	{"mai-code-1.1-flash", "/responses"},
	35	+	{"gemini-3.8-flash", "/chat/completions"},
	36	+	{"kimi-k3", "/chat/completions"},
	37	+	} {
	38	+	t.Run(tc.model, func(t *testing.T) {
	39	+	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
	40	+	if r.URL.Path != tc.path {
	41	+	t.Errorf("path = %q, want %q", r.URL.Path, tc.path)
	42	+	}
	43	+	if r.Header.Get("Authorization") != "Bearer synthetic-session-token" {
	44	+	t.Error("missing session Bearer auth")
	45	+	}
	46	+	if r.Header.Get("Copilot-Integration-Id") != "vscode-chat" \|\| r.Header.Get("X-Initiator") != "user" \|\| r.Header.Get("Openai-Intent") != "conversation-edits" {
	47	+	t.Error("missing Copilot identity headers")
	48	+	}
	49	+	for _, header := range []string{"x-api-key", "anthropic-beta", "x-app", "chatgpt-account-id", "openai-beta", "originator"} {
	50	+	if r.Header.Get(header) != "" {
	51	+	t.Errorf("unexpected identity header %s", header)
	52	+	}
	53	+	}
	54	+	var body map[string]json.RawMessage
	55	+	if err := json.NewDecoder(r.Body).Decode(&body); err != nil {
	56	+	t.Error(err)
	57	+	w.WriteHeader(400)
	58	+	return
	59	+	}
	60	+	if string(body["model"]) != fmt.Sprintf("%q", tc.model) {
	61	+	t.Error("model ID changed on wire")
	62	+	}
	63	+	w.Header().Set("Content-Type", "text/event-stream")
	64	+	switch tc.path {
	65	+	case "/v1/messages":
	66	+	if len(body["messages"]) == 0 \|\| len(body["max_tokens"]) == 0 {
	67	+	t.Error("missing Messages fields")
	68	+	}
	69	+	if strings.Contains(string(body["system"]), "Claude Code") {
	70	+	t.Error("injected Claude Code system identity")
	71	+	}
	72	+	if tc.model == "claude-sonnet-5" {
	73	+	if string(body["thinking"]) != `{"type":"adaptive"}` {
	74	+	t.Errorf("thinking = %s", body["thinking"])
	75	+	}
	76	+	if len(body["output_config"]) == 0 \|\| len(body["temperature"]) != 0 {
	77	+	t.Error("invalid adaptive thinking payload")
	78	+	}
	79	+	}
	80	+	fmt.Fprint(w, "event: message_start\ndata: {\"message\":{\"id\":\"test\",\"usage\":{\"input_tokens\":1}}}\n\nevent: message_delta\ndata: {\"delta\":{\"stop_reason\":\"end_turn\"},\"usage\":{\"output_tokens\":1}}\n\nevent: message_stop\ndata: {}\n\n")
	81	+	case "/responses":
	82	+	if len(body["input"]) == 0 \|\| len(body["reasoning"]) == 0 {
	83	+	t.Error("missing Responses fields")
	84	+	}
	85	+	fmt.Fprint(w, "data: {\"type\":\"response.completed\",\"response\":{\"status\":\"completed\",\"usage\":{\"input_tokens\":1,\"output_tokens\":1}}}\n\n")
	86	+	default:
	87	+	if len(body["reasoning_effort"]) != 0 {
	88	+	t.Error("unsupported reasoning_effort sent")
	89	+	}
	90	+	if len(body["messages"]) == 0 {
	91	+	t.Error("missing chat messages")
	92	+	}
	93	+	fmt.Fprint(w, "data: {\"choices\":[{\"index\":0,\"delta\":{\"content\":\"ok\"},\"finish_reason\":\"stop\"}]}\n\ndata: [DONE]\n\n")
	94	+	}
	95	+	}))
	96	+	defer srv.Close()
	97	+	client := NewGithubCopilotClient(seedCopilotTestToken(t, srv.URL))
	98	+	stream, err := client.Stream(context.Background(), Request{Model: tc.model, Reasoning: "high", System: "Test system", Messages: []Message{{Role: RoleUser, Content: []Content{TextBlock{Text: "Hello"}}}}})
	99	+	if err != nil {
	100	+	t.Fatal(err)
	101	+	}
	102	+	for event := range stream {
	103	+	if e, ok := event.(EventDone); ok && e.Err != nil {
	104	+	t.Fatalf("stream error: %v", e.Err)
	105	+	}
	106	+	if e, ok := event.(EventStart); ok && e.Provider != "github-copilot" {
	107	+	t.Errorf("event provider = %s", e.Provider)
	108	+	}
	109	+	}
	110	+	})
	111	+	}
	112	+	}
	113	+
	114	+	type copilotMetadataCapture struct {
	115	+	metadata copilotRequestMetadata
	116	+	reasoning string
	117	+	}
	118	+
	119	+	func (c *copilotMetadataCapture) Name() string { return "github-copilot" }
	120	+	func (c *copilotMetadataCapture) Stream(ctx context.Context, req Request) (<-chan Event, error) {
	121	+	c.metadata = ctx.Value(copilotRequestKey{}).(copilotRequestMetadata)
	122	+	c.reasoning = req.Reasoning
	123	+	ch := make(chan Event)
	124	+	close(ch)
	125	+	return ch, nil
	126	+	}
	127	+
	128	+	func TestCopilotDynamicRequestMetadata(t *testing.T) {
	129	+	client := NewGithubCopilotClient("unused").(*copilotClient)
	130	+	capture := &copilotMetadataCapture{}
	131	+	client.router.byAPI[APICompletions] = capture
	132	+	for _, tc := range []struct {
	133	+	messages []Message
	134	+	initiator string
	135	+	vision bool
	136	+	}{
	137	+	{nil, "user", false},
	138	+	{[]Message{{Role: RoleUser, Content: []Content{ImageBlock{}}}}, "user", true},
	139	+	{[]Message{{Role: RoleTool, Content: []Content{ToolResultBlock{Content: []Content{ImageBlock{}}}}}}, "agent", true},
	140	+	{[]Message{{Role: RoleAssistant}}, "agent", false},
	141	+	{[]Message{{Role: RoleUser}}, "user", false},
	142	+	} {
	143	+	_, err := client.Stream(context.Background(), Request{Model: "kimi-k3", Reasoning: "high", Messages: tc.messages})
	144	+	if err != nil {
	145	+	t.Fatal(err)
	146	+	}
	147	+	if capture.metadata.initiator != tc.initiator \|\| capture.metadata.vision != tc.vision \|\| capture.reasoning != "" {
	148	+	t.Errorf("unexpected metadata: %+v, reasoning %q", capture.metadata, capture.reasoning)
	149	+	}
	150	+	}
	151	+	// The transport must propagate vision metadata and remove it on the next text-only request.
	152	+	mock := &mockRoundTripper{}
	153	+	transport := &copilotRefreshTransport{inner: mock, pat: seedCopilotTestToken(t, copilotDefaultBaseURL)}
	154	+	for _, vision := range []bool{true, false} {
	155	+	ctx := context.WithValue(context.Background(), copilotRequestKey{}, copilotRequestMetadata{initiator: "agent", vision: vision})
	156	+	req, _ := http.NewRequestWithContext(ctx, "POST", copilotDefaultBaseURL+"/responses", nil)
	157	+	req.Header.Set("Copilot-Vision-Request", "stale")
	158	+	resp, err := transport.RoundTrip(req)
	159	+	if err != nil {
	160	+	t.Fatal(err)
	161	+	}
	162	+	resp.Body.Close()
	163	+	want := ""
	164	+	if vision {
	165	+	want = "true"
	166	+	}
	167	+	if mock.lastReq.Header.Get("Copilot-Vision-Request") != want \|\| mock.lastReq.Header.Get("X-Initiator") != "agent" {
	168	+	t.Error("incorrect dynamic transport headers")
	169	+	}
	170	+	if req.Header.Get("Copilot-Vision-Request") != "stale" {
	171	+	t.Error("transport mutated caller headers")
	172	+	}
	173	+	}
	174	+	}

M packages/provider/gpt6_astra_test.go +3 -3


@@ -35,12 +35,12 @@ func TestGPT6AstraCatalog(t *testing.T) {
35	35		}
36	36
37	37		func TestGPT6AstraResponsesReasoning(t *testing.T) {
38		-	copilot := NewGithubCopilotClient("test-token").(*modelRouter)
	38	+	copilot := NewGithubCopilotClient("test-token").(*copilotClient).router
39	39		clients := map[string]*codexClient{
40	40		"openai": NewOpenAIResponsesNamed("test-token", "", "openai").(renamedClient).inner.(codexClient),
41	41		"openai-responses": NewOpenAIResponsesNamed("test-token", "", "openai-responses").(renamedClient).inner.(codexClient),
42	42		"openai-codex": NewOpenAICodex("test-token", "test-account", "").(*codexClient),
43		-	"github-copilot": copilot.byAPI[APIResponses].(renamedClient).inner.(codexClient),
	43	+	"github-copilot": copilot.byAPI[APIResponses].(*codexClient),
44	44		}
45	45		for name, client := range clients {
46	46		t.Run(name, func(t *testing.T) {

@@ -65,7 +65,7 @@ func TestGPT6AstraResponsesReasoning(t *testing.T) {
65	65		}
66	66
67	67		func TestGPT6AstraCopilotDispatch(t *testing.T) {
68		-	router := NewGithubCopilotClient("test-token").(*modelRouter)
	68	+	router := NewGithubCopilotClient("test-token").(*copilotClient).router
69	69		completions := &routeCaptureClient{name: "github-copilot"}
70	70		responses := &routeCaptureClient{name: "github-copilot"}
71	71		router.fallback = completions


@@ -35,12 +35,12 @@ func TestGPT6AstraCatalog(t *testing.T) {
35		}	35		}
36			36
37		func TestGPT6AstraResponsesReasoning(t *testing.T) {	37		func TestGPT6AstraResponsesReasoning(t *testing.T) {
38	-	copilot := NewGithubCopilotClient("test-token").(*modelRouter)	38	+	copilot := NewGithubCopilotClient("test-token").(*copilotClient).router
39		clients := map[string]*codexClient{	39		clients := map[string]*codexClient{
40		"openai": NewOpenAIResponsesNamed("test-token", "", "openai").(renamedClient).inner.(codexClient),	40		"openai": NewOpenAIResponsesNamed("test-token", "", "openai").(renamedClient).inner.(codexClient),
41		"openai-responses": NewOpenAIResponsesNamed("test-token", "", "openai-responses").(renamedClient).inner.(codexClient),	41		"openai-responses": NewOpenAIResponsesNamed("test-token", "", "openai-responses").(renamedClient).inner.(codexClient),
42		"openai-codex": NewOpenAICodex("test-token", "test-account", "").(*codexClient),	42		"openai-codex": NewOpenAICodex("test-token", "test-account", "").(*codexClient),
43	-	"github-copilot": copilot.byAPI[APIResponses].(renamedClient).inner.(codexClient),	43	+	"github-copilot": copilot.byAPI[APIResponses].(*codexClient),
44		}	44		}
45		for name, client := range clients {	45		for name, client := range clients {
46		t.Run(name, func(t *testing.T) {	46		t.Run(name, func(t *testing.T) {

@@ -65,7 +65,7 @@ func TestGPT6AstraResponsesReasoning(t *testing.T) {
65		}	65		}
66			66
67		func TestGPT6AstraCopilotDispatch(t *testing.T) {	67		func TestGPT6AstraCopilotDispatch(t *testing.T) {
68	-	router := NewGithubCopilotClient("test-token").(*modelRouter)	68	+	router := NewGithubCopilotClient("test-token").(*copilotClient).router
69		completions := &routeCaptureClient{name: "github-copilot"}	69		completions := &routeCaptureClient{name: "github-copilot"}
70		responses := &routeCaptureClient{name: "github-copilot"}	70		responses := &routeCaptureClient{name: "github-copilot"}
71		router.fallback = completions	71		router.fallback = completions

M packages/provider/model_router.go +2 -0


@@ -6,6 +6,8 @@ import (
6	6		)
7	7
8	8		const (
	9	+	// APIAnthropicMessages identifies the Anthropic Messages wire API.
	10	+	APIAnthropicMessages = "anthropic-messages"
9	11		// APICompletions identifies the OpenAI Chat Completions wire API.
10	12		APICompletions = "openai-completions"
11	13		// APIResponses identifies the OpenAI Responses wire API.


@@ -6,6 +6,8 @@ import (
6	)	6		)
7		7
8	const (	8		const (
		9	+	// APIAnthropicMessages identifies the Anthropic Messages wire API.
		10	+	APIAnthropicMessages = "anthropic-messages"
9	// APICompletions identifies the OpenAI Chat Completions wire API.	11		// APICompletions identifies the OpenAI Chat Completions wire API.
10	APICompletions = "openai-completions"	12		APICompletions = "openai-completions"
11	// APIResponses identifies the OpenAI Responses wire API.	13		// APIResponses identifies the OpenAI Responses wire API.

M packages/provider/models.go +46 -7


@@ -449,10 +449,11 @@ var DefaultModel = Catalog[0] // claude-sonnet-4-5
449	449		// models loaded via SetLiveModels.
450	450
451	451		var (
452		-	activeMu sync.RWMutex
453		-	active []Model // live overlay merged in via SetLiveModels; nil = none yet
454		-	activeSet bool // true once SetLiveModels has run (even with empty live)
455		-	managedModels []Model // ephemeral models exposed by local model managers
	452	+	activeMu sync.RWMutex
	453	+	active []Model // live overlay merged in via SetLiveModels; nil = none yet
	454	+	activeSet bool // true once SetLiveModels has run (even with empty live)
	455	+	managedModels []Model // ephemeral models exposed by local model managers
	456	+	modelAvailability = map[string]map[string]bool{} // account-specific, memory only
456	457		)
457	458
458	459		// SetLiveModels replaces the "live" overlay used by the active catalog.

@@ -479,7 +480,9 @@ func SetLiveModels(live []Model) {
479	480		// Snapshotting Catalog at var-init time would freeze the picker to the
480	481		// curated seed list and drop every extra provider (openrouter, groq,
481	482		// xai, ...). Deferring the read to call time avoids that ordering trap.
482		-	func Active() []Model {
	483	+	func Active() []Model { return activeModels(true) }
	484	+
	485	+	func activeModels(availableOnly bool) []Model {
483	486		activeMu.RLock()
484	487		defer activeMu.RUnlock()
485	488		src := active

@@ -492,6 +495,9 @@ func Active() []Model {
492	495		out[i].ReasoningLevelMap = maps.Clone(out[i].ReasoningLevelMap)
493	496		}
494	497		if len(managedModels) == 0 {
	498	+	if availableOnly {
	499	+	return filterAvailableModels(out)
	500	+	}
495	501		return out
496	502		}
497	503		index := make(map[string]int, len(out))

@@ -508,6 +514,38 @@ func Active() []Model {
508	514		index[key] = len(out)
509	515		out = append(out, model)
510	516		}
	517	+	if availableOnly {
	518	+	return filterAvailableModels(out)
	519	+	}
	520	+	return out
	521	+	}
	522	+
	523	+	// SetModelAvailability restricts a provider's visible catalog to the supplied
	524	+	// account model IDs. A nil slice clears the restriction; an empty non-nil slice
	525	+	// hides all models. This snapshot is not persisted in the shared model cache.
	526	+	func SetModelAvailability(provider string, ids []string) {
	527	+	activeMu.Lock()
	528	+	defer activeMu.Unlock()
	529	+	if ids == nil {
	530	+	delete(modelAvailability, provider)
	531	+	return
	532	+	}
	533	+	allowed := make(map[string]bool, len(ids))
	534	+	for _, id := range ids {
	535	+	allowed[id] = true
	536	+	}
	537	+	modelAvailability[provider] = allowed
	538	+	}
	539	+
	540	+	// Caller holds activeMu. Models is already a private copy.
	541	+	func filterAvailableModels(models []Model) []Model {
	542	+	out := models[:0]
	543	+	for _, m := range models {
	544	+	allowed, restricted := modelAvailability[m.Provider]
	545	+	if !restricted \|\| allowed[m.ID] {
	546	+	out = append(out, m)
	547	+	}
	548	+	}
511	549		return out
512	550		}
513	551

@@ -522,9 +560,10 @@ func SetManagedModels(models []Model) {
522	560
523	561		// FindModel returns a Model by id, optionally constrained by provider.
524	562		// If provider is empty, the first matching id is returned. Looks up
525		-	// against the merged active catalog.
	563	+	// against the merged catalog, including models hidden by account availability.
	564	+	// Existing sessions still need stable wire and pricing metadata after a refresh.
526	565		func FindModel(provider, id string) (Model, error) {
527		-	for _, m := range Active() {
	566	+	for _, m := range activeModels(false) {
528	567		if m.ID == id && (provider == "" \|\| m.Provider == provider) {
529	568		return m, nil
530	569		}


@@ -449,10 +449,11 @@ var DefaultModel = Catalog[0] // claude-sonnet-4-5
449		// models loaded via SetLiveModels.	449		// models loaded via SetLiveModels.
450			450
451		var (	451		var (
452	-	activeMu sync.RWMutex	452	+	activeMu sync.RWMutex
453	-	active []Model // live overlay merged in via SetLiveModels; nil = none yet	453	+	active []Model // live overlay merged in via SetLiveModels; nil = none yet
454	-	activeSet bool // true once SetLiveModels has run (even with empty live)	454	+	activeSet bool // true once SetLiveModels has run (even with empty live)
455	-	managedModels []Model // ephemeral models exposed by local model managers	455	+	managedModels []Model // ephemeral models exposed by local model managers
			456	+	modelAvailability = map[string]map[string]bool{} // account-specific, memory only
456		)	457		)
457			458
458		// SetLiveModels replaces the "live" overlay used by the active catalog.	459		// SetLiveModels replaces the "live" overlay used by the active catalog.

@@ -479,7 +480,9 @@ func SetLiveModels(live []Model) {
479		// Snapshotting Catalog at var-init time would freeze the picker to the	480		// Snapshotting Catalog at var-init time would freeze the picker to the
480		// curated seed list and drop every extra provider (openrouter, groq,	481		// curated seed list and drop every extra provider (openrouter, groq,
481		// xai, ...). Deferring the read to call time avoids that ordering trap.	482		// xai, ...). Deferring the read to call time avoids that ordering trap.
482	-	func Active() []Model {	483	+	func Active() []Model { return activeModels(true) }
			484	+
			485	+	func activeModels(availableOnly bool) []Model {
483		activeMu.RLock()	486		activeMu.RLock()
484		defer activeMu.RUnlock()	487		defer activeMu.RUnlock()
485		src := active	488		src := active

@@ -492,6 +495,9 @@ func Active() []Model {
492		out[i].ReasoningLevelMap = maps.Clone(out[i].ReasoningLevelMap)	495		out[i].ReasoningLevelMap = maps.Clone(out[i].ReasoningLevelMap)
493		}	496		}
494		if len(managedModels) == 0 {	497		if len(managedModels) == 0 {
			498	+	if availableOnly {
			499	+	return filterAvailableModels(out)
			500	+	}
495		return out	501		return out
496		}	502		}
497		index := make(map[string]int, len(out))	503		index := make(map[string]int, len(out))

@@ -508,6 +514,38 @@ func Active() []Model {
508		index[key] = len(out)	514		index[key] = len(out)
509		out = append(out, model)	515		out = append(out, model)
510		}	516		}
			517	+	if availableOnly {
			518	+	return filterAvailableModels(out)
			519	+	}
			520	+	return out
			521	+	}
			522	+
			523	+	// SetModelAvailability restricts a provider's visible catalog to the supplied
			524	+	// account model IDs. A nil slice clears the restriction; an empty non-nil slice
			525	+	// hides all models. This snapshot is not persisted in the shared model cache.
			526	+	func SetModelAvailability(provider string, ids []string) {
			527	+	activeMu.Lock()
			528	+	defer activeMu.Unlock()
			529	+	if ids == nil {
			530	+	delete(modelAvailability, provider)
			531	+	return
			532	+	}
			533	+	allowed := make(map[string]bool, len(ids))
			534	+	for _, id := range ids {
			535	+	allowed[id] = true
			536	+	}
			537	+	modelAvailability[provider] = allowed
			538	+	}
			539	+
			540	+	// Caller holds activeMu. Models is already a private copy.
			541	+	func filterAvailableModels(models []Model) []Model {
			542	+	out := models[:0]
			543	+	for _, m := range models {
			544	+	allowed, restricted := modelAvailability[m.Provider]
			545	+	if !restricted \|\| allowed[m.ID] {
			546	+	out = append(out, m)
			547	+	}
			548	+	}
511		return out	549		return out
512		}	550		}
513			551

@@ -522,9 +560,10 @@ func SetManagedModels(models []Model) {
522			560
523		// FindModel returns a Model by id, optionally constrained by provider.	561		// FindModel returns a Model by id, optionally constrained by provider.
524		// If provider is empty, the first matching id is returned. Looks up	562		// If provider is empty, the first matching id is returned. Looks up
525	-	// against the merged active catalog.	563	+	// against the merged catalog, including models hidden by account availability.
			564	+	// Existing sessions still need stable wire and pricing metadata after a refresh.
526		func FindModel(provider, id string) (Model, error) {	565		func FindModel(provider, id string) (Model, error) {
527	-	for _, m := range Active() {	566	+	for _, m := range activeModels(false) {
528		if m.ID == id && (provider == "" \|\| m.Provider == provider) {	567		if m.ID == id && (provider == "" \|\| m.Provider == provider) {
529		return m, nil	568		return m, nil
530		}	569		}

M packages/provider/openai_url_test.go +3 -3


@@ -40,11 +40,11 @@ func TestChatCompletionsURL(t *testing.T) {
40	40		}
41	41
42	42		func TestGithubCopilotChatCompletionsURL(t *testing.T) {
43		-	router, ok := NewGithubCopilotClient("test-token").(*modelRouter)
	43	+	copilot, ok := NewGithubCopilotClient("test-token").(*copilotClient)
44	44		if !ok {
45		-	t.Fatal("NewGithubCopilotClient did not return a modelRouter")
	45	+	t.Fatal("NewGithubCopilotClient did not return a copilotClient")
46	46		}
47		-	client, ok := router.fallback.(*openaiClient)
	47	+	client, ok := copilot.router.fallback.(*openaiClient)
48	48		if !ok {
49	49		t.Fatal("GitHub Copilot router fallback is not an openaiClient")
50	50		}


@@ -40,11 +40,11 @@ func TestChatCompletionsURL(t *testing.T) {
40		}	40		}
41			41
42		func TestGithubCopilotChatCompletionsURL(t *testing.T) {	42		func TestGithubCopilotChatCompletionsURL(t *testing.T) {
43	-	router, ok := NewGithubCopilotClient("test-token").(*modelRouter)	43	+	copilot, ok := NewGithubCopilotClient("test-token").(*copilotClient)
44		if !ok {	44		if !ok {
45	-	t.Fatal("NewGithubCopilotClient did not return a modelRouter")	45	+	t.Fatal("NewGithubCopilotClient did not return a copilotClient")
46		}	46		}
47	-	client, ok := router.fallback.(*openaiClient)	47	+	client, ok := copilot.router.fallback.(*openaiClient)
48		if !ok {	48		if !ok {
49		t.Fatal("GitHub Copilot router fallback is not an openaiClient")	49		t.Fatal("GitHub Copilot router fallback is not an openaiClient")
50		}	50		}