1package provider
2
3// Amazon Bedrock Converse-Stream client.
4//
5// Endpoint: POST https://bedrock-runtime.{region}.amazonaws.com/model/{id}/converse-stream
6//
7// Auth: two paths supported.
8//
9// - **Bearer token** via AWS_BEARER_TOKEN_BEDROCK. Modern, simple,
10// `bedrock:CallWithBearerToken` IAM permission required.
11// - **SigV4 signing** via AWS_ACCESS_KEY_ID + AWS_SECRET_ACCESS_KEY
12// (+ optional AWS_SESSION_TOKEN for temporary credentials, e.g. SSO
13// or IRSA-issued). Hand-rolled here so we don't pull in aws-sdk-go-v2.
14//
15// If both are present, bearer wins (simpler / fewer moving parts).
16//
17// Wire format: AWS event-stream binary framing. Each message is:
18//
19// [4-byte total length BE]
20// [4-byte headers length BE]
21// [4-byte CRC32 of the 8-byte prelude]
22// [headers ...] each = [1B name-length][name][1B value-type][2B value-length][value]
23// [payload ...] payload-length = total - headers - 4 (trailer) - 12 (prelude+prelude-crc)
24// [4-byte CRC32 of message bytes minus trailer]
25//
26// We only consume the stream — never produce it — so we don't validate
27// CRCs (the TLS+TCP transport already protects in-flight bytes). We do
28// validate length-bounds to avoid OOB reads on a hostile / corrupted
29// stream.
30//
31// Event types we care about (from `:event-type` header):
32// - messageStart { role: "assistant" }
33// - contentBlockStart { contentBlockIndex, start: { toolUse: { toolUseId, name } } }
34// - contentBlockDelta { contentBlockIndex, delta: { text?, toolUse?: { input } } }
35// - contentBlockStop { contentBlockIndex }
36// - messageStop { stopReason }
37// - metadata { usage: { inputTokens, outputTokens } }
38
39import (
40 "bytes"
41 "context"
42 "crypto/hmac"
43 "crypto/sha256"
44 "encoding/base64"
45 "encoding/binary"
46 "encoding/hex"
47 "encoding/json"
48 "fmt"
49 "io"
50 "net/http"
51 "net/url"
52 "os"
53 "sort"
54 "strings"
55 "time"
56)
57
58type bedrockClient struct {
59 // Auth mode is determined at construction time. Exactly one of
60 // bearerToken or sigv4 is populated.
61 bearerToken string
62 sigv4 *bedrockSigV4Creds
63
64 region string
65 baseURL string
66 http *http.Client
67}
68
69// bedrockSigV4Creds holds the IAM credentials used to sign each request.
70type bedrockSigV4Creds struct {
71 accessKeyID string
72 secretAccessKey string
73 sessionToken string // optional; STS / SSO / IRSA temp creds
74}
75
76// NewBedrockClient returns a Bedrock client.
77//
78// Auth resolution (first match wins):
79//
80// 1. apiKey == real bearer-ish string (not "<aws>") -> bearer route.
81// 2. AWS_BEARER_TOKEN_BEDROCK env var -> bearer route.
82// 3. AWS_ACCESS_KEY_ID + AWS_SECRET_ACCESS_KEY (+ optional
83// AWS_SESSION_TOKEN) -> SigV4 route.
84// 4. AWS_PROFILE -> read ~/.aws/credentials, take that profile's keys.
85//
86// region defaults to us-east-1 unless AWS_REGION / AWS_DEFAULT_REGION is
87// set or the baseURL embeds a region.
88func NewBedrockClient(apiKey, baseURL string) Client {
89 region := os.Getenv("AWS_REGION")
90 if region == "" {
91 region = os.Getenv("AWS_DEFAULT_REGION")
92 }
93 if region == "" {
94 region = "us-east-1"
95 }
96 if baseURL == "" {
97 baseURL = "https://bedrock-runtime." + region + ".amazonaws.com"
98 }
99 c := &bedrockClient{
100 region: region,
101 baseURL: strings.TrimRight(baseURL, "/"),
102 http: &http.Client{Timeout: 0},
103 }
104
105 // Bearer route.
106 token := apiKey
107 if token == "" || token == "<aws>" {
108 token = os.Getenv("AWS_BEARER_TOKEN_BEDROCK")
109 }
110 if token != "" && token != "<aws>" {
111 c.bearerToken = token
112 return c
113 }
114
115 // SigV4 route: env vars first.
116 ak := os.Getenv("AWS_ACCESS_KEY_ID")
117 sk := os.Getenv("AWS_SECRET_ACCESS_KEY")
118 st := os.Getenv("AWS_SESSION_TOKEN")
119 if ak != "" && sk != "" {
120 c.sigv4 = &bedrockSigV4Creds{accessKeyID: ak, secretAccessKey: sk, sessionToken: st}
121 return c
122 }
123
124 // SigV4 route: ~/.aws/credentials via AWS_PROFILE.
125 if profile := os.Getenv("AWS_PROFILE"); profile != "" {
126 if creds, err := readAWSCredentialsFile(profile); err == nil {
127 c.sigv4 = creds
128 return c
129 }
130 }
131
132 return &unimplementedClient{
133 name: "amazon-bedrock",
134 hint: "no Bedrock credentials found (set AWS_BEARER_TOKEN_BEDROCK, AWS_ACCESS_KEY_ID+AWS_SECRET_ACCESS_KEY, or AWS_PROFILE)",
135 }
136}
137
138// readAWSCredentialsFile parses ~/.aws/credentials and returns the
139// access-key/secret-key (and optional session-token) for the named
140// profile. The file is an INI-like format with `[profile]` headers.
141func readAWSCredentialsFile(profile string) (*bedrockSigV4Creds, error) {
142 home, err := os.UserHomeDir()
143 if err != nil {
144 return nil, err
145 }
146 b, err := os.ReadFile(home + "/.aws/credentials")
147 if err != nil {
148 return nil, err
149 }
150 var current string
151 creds := map[string]map[string]string{}
152 for _, line := range strings.Split(string(b), "\n") {
153 line = strings.TrimSpace(line)
154 if line == "" || strings.HasPrefix(line, "#") {
155 continue
156 }
157 if strings.HasPrefix(line, "[") && strings.HasSuffix(line, "]") {
158 current = strings.TrimSpace(line[1 : len(line)-1])
159 creds[current] = map[string]string{}
160 continue
161 }
162 if current == "" {
163 continue
164 }
165 eq := strings.IndexByte(line, '=')
166 if eq < 0 {
167 continue
168 }
169 k := strings.TrimSpace(line[:eq])
170 v := strings.TrimSpace(line[eq+1:])
171 creds[current][k] = v
172 }
173 p, ok := creds[profile]
174 if !ok {
175 return nil, fmt.Errorf("aws profile %q not found in ~/.aws/credentials", profile)
176 }
177 ak := p["aws_access_key_id"]
178 sk := p["aws_secret_access_key"]
179 if ak == "" || sk == "" {
180 return nil, fmt.Errorf("aws profile %q missing aws_access_key_id or aws_secret_access_key", profile)
181 }
182 return &bedrockSigV4Creds{
183 accessKeyID: ak,
184 secretAccessKey: sk,
185 sessionToken: p["aws_session_token"],
186 }, nil
187}
188
189func (c *bedrockClient) Name() string { return "amazon-bedrock" }
190
191// ---- request building ----
192
193type bedrockMessage struct {
194 Role string `json:"role"`
195 Content []map[string]interface{} `json:"content"`
196}
197
198// bedrockCachePoint is the Converse API cache checkpoint content block.
199// Appending this to a system or message content array tells Bedrock to
200// create a cache boundary at that position in the prompt prefix.
201var bedrockCachePoint = map[string]interface{}{
202 "cachePoint": map[string]interface{}{"type": "default"},
203}
204
205type bedrockToolSpec struct {
206 ToolSpec struct {
207 Name string `json:"name"`
208 Description string `json:"description,omitempty"`
209 InputSchema struct {
210 JSON json.RawMessage `json:"json"`
211 } `json:"inputSchema"`
212 } `json:"toolSpec"`
213}
214
215type bedrockRequest struct {
216 Messages []bedrockMessage `json:"messages"`
217 System []map[string]interface{} `json:"system,omitempty"`
218 InferenceConfig struct {
219 MaxTokens int `json:"maxTokens,omitempty"`
220 Temperature *float32 `json:"temperature,omitempty"`
221 } `json:"inferenceConfig,omitempty"`
222 ToolConfig *struct {
223 Tools []bedrockToolSpec `json:"tools"`
224 } `json:"toolConfig,omitempty"`
225}
226
227func normalizeBedrockToolResults(msgs []Message) []Message {
228 resultByID := map[string]ToolResultBlock{}
229 for _, m := range msgs {
230 for _, c := range m.Content {
231 if tr, ok := c.(ToolResultBlock); ok {
232 if _, exists := resultByID[tr.CallID]; !exists {
233 resultByID[tr.CallID] = tr
234 }
235 }
236 }
237 }
238
239 out := make([]Message, 0, len(msgs))
240 for _, m := range msgs {
241 copy := m
242 copy.Content = nil
243 var toolCalls []ToolCallBlock
244 for _, c := range m.Content {
245 switch v := c.(type) {
246 case ToolResultBlock:
247 // Bedrock requires toolResult blocks immediately after the
248 // assistant toolUse they answer. Reinsert them from the
249 // assistant pass below instead of preserving their original
250 // location, which may be separated by user text in active
251 // sessions.
252 continue
253 case ToolCallBlock:
254 toolCalls = append(toolCalls, v)
255 }
256 copy.Content = append(copy.Content, c)
257 }
258 if len(copy.Content) > 0 {
259 out = append(out, copy)
260 }
261 if m.Role != RoleAssistant || len(toolCalls) == 0 {
262 continue
263 }
264
265 results := make([]Content, 0, len(toolCalls))
266 for _, tc := range toolCalls {
267 if tr, ok := resultByID[tc.ID]; ok {
268 results = append(results, tr)
269 continue
270 }
271 results = append(results, ToolResultBlock{
272 CallID: tc.ID,
273 IsError: true,
274 Content: []Content{TextBlock{Text: "tool call did not complete before the next user message"}},
275 })
276 }
277 out = append(out, Message{Role: RoleTool, Content: results, Time: m.Time})
278 }
279 return out
280}
281
282// bedrockModelSupportsCaching reports whether the resolved model ID
283// supports explicit prompt caching via cachePoint markers on Bedrock.
284// We use PriceCacheWrite > 0 as a proxy: every Bedrock-hosted Claude
285// model with a write price in the catalog supports cachePoint markers.
286// Nova models use automatic caching and don't need explicit markers.
287func bedrockModelSupportsCaching(modelID string) bool {
288 // Strip geo prefix (us./eu./apac./au./global.) before catalog lookup.
289 for _, p := range bedrockGeoPrefixes {
290 if strings.HasPrefix(modelID, p+".") {
291 modelID = modelID[len(p)+1:]
292 break
293 }
294 }
295 if m, err := FindModel("amazon-bedrock", modelID); err == nil {
296 return m.PriceCacheWrite > 0
297 }
298 // Unknown model: enable for Anthropic Claude families — cachePoint is
299 // silently ignored by the API if the model doesn't support it.
300 return strings.HasPrefix(modelID, "anthropic.claude-")
301}
302
303// bedrockMessagesHaveToolBlocks reports whether any message in msgs
304// contains a toolUse or toolResult content block. Bedrock's Converse API
305// rejects the request with HTTP 400 if those block types are present but
306// toolConfig is absent.
307func bedrockMessagesHaveToolBlocks(msgs []bedrockMessage) bool {
308 for _, m := range msgs {
309 for _, c := range m.Content {
310 if _, ok := c["toolUse"]; ok {
311 return true
312 }
313 if _, ok := c["toolResult"]; ok {
314 return true
315 }
316 }
317 }
318 return false
319}
320
321func (c *bedrockClient) buildRequest(req Request) (*bedrockRequest, error) {
322 out := &bedrockRequest{}
323
324 // Resolve the model ID as it will appear on the wire so the caching
325 // check operates on the same ID used for FindModel.
326 resolvedModel := resolveBedrockInferenceProfileID(req.Model, c.region)
327 caching := bedrockModelSupportsCaching(resolvedModel)
328
329 if req.System != "" {
330 sysBlock := map[string]interface{}{"text": req.System}
331 if caching {
332 // Append cachePoint after the system text so the stable system
333 // prompt is cached as the first breakpoint.
334 out.System = []map[string]interface{}{sysBlock, bedrockCachePoint}
335 } else {
336 out.System = []map[string]interface{}{sysBlock}
337 }
338 }
339 out.InferenceConfig.Temperature = req.Temperature
340 out.InferenceConfig.MaxTokens = req.MaxTokens
341 if out.InferenceConfig.MaxTokens == 0 {
342 out.InferenceConfig.MaxTokens = 4096
343 }
344 for _, m := range normalizeBedrockToolResults(req.Messages) {
345 role := string(m.Role)
346 if role == "tool" {
347 role = "user"
348 }
349 bm := bedrockMessage{Role: role}
350 for _, c := range m.Content {
351 switch v := c.(type) {
352 case TextBlock:
353 bm.Content = append(bm.Content, map[string]interface{}{"text": v.Text})
354 case ToolCallBlock:
355 var input map[string]interface{}
356 _ = json.Unmarshal(v.Arguments, &input)
357 if input == nil {
358 input = map[string]interface{}{}
359 }
360 bm.Content = append(bm.Content, map[string]interface{}{
361 "toolUse": map[string]interface{}{
362 "toolUseId": v.ID, "name": v.Name, "input": input,
363 },
364 })
365 case ImageBlock:
366 data, mime := anthShrinkImageBytesIfTooBig(v.Data, v.MimeType)
367 // Bedrock Converse uses the bare format name (jpeg/png/gif/webp).
368 imgFmt := strings.TrimPrefix(mime, "image/")
369 bm.Content = append(bm.Content, map[string]interface{}{
370 "image": map[string]interface{}{
371 "format": imgFmt,
372 "source": map[string]interface{}{
373 "bytes": base64.StdEncoding.EncodeToString(data),
374 },
375 },
376 })
377 case ToolResultBlock:
378 var resultContent []map[string]interface{}
379 for _, inner := range v.Content {
380 switch iv := inner.(type) {
381 case TextBlock:
382 resultContent = append(resultContent, map[string]interface{}{"text": iv.Text})
383 case ImageBlock:
384 data, mime := anthShrinkImageBytesIfTooBig(iv.Data, iv.MimeType)
385 imgFmt := strings.TrimPrefix(mime, "image/")
386 resultContent = append(resultContent, map[string]interface{}{
387 "image": map[string]interface{}{
388 "format": imgFmt,
389 "source": map[string]interface{}{
390 "bytes": base64.StdEncoding.EncodeToString(data),
391 },
392 },
393 })
394 }
395 }
396 status := "success"
397 if v.IsError {
398 status = "error"
399 }
400 bm.Content = append(bm.Content, map[string]interface{}{
401 "toolResult": map[string]interface{}{
402 "toolUseId": v.CallID,
403 "content": resultContent,
404 "status": status,
405 },
406 })
407 }
408 }
409 if len(bm.Content) == 0 {
410 continue
411 }
412 out.Messages = append(out.Messages, bm)
413 }
414 if len(req.Tools) > 0 {
415 tc := struct {
416 Tools []bedrockToolSpec `json:"tools"`
417 }{}
418 for _, t := range req.Tools {
419 var ts bedrockToolSpec
420 ts.ToolSpec.Name = t.Name
421 ts.ToolSpec.Description = t.Description
422 ts.ToolSpec.InputSchema.JSON = t.Schema
423 tc.Tools = append(tc.Tools, ts)
424 }
425 out.ToolConfig = &tc
426 } else if bedrockMessagesHaveToolBlocks(out.Messages) {
427 // Bedrock requires toolConfig to be present whenever the message
428 // history contains toolUse or toolResult content blocks, even if
429 // the current request does not intend to call any tools (e.g. the
430 // /btw side-chat sends the frozen main transcript which may include
431 // prior tool turns). Inject a stub tool so the API accepts the
432 // request; it will never be invoked since no tool_use stop reason
433 // can be triggered when the real tool list is empty.
434 var stub bedrockToolSpec
435 stub.ToolSpec.Name = "_placeholder"
436 stub.ToolSpec.Description = "placeholder required by Bedrock when tool history is present"
437 stub.ToolSpec.InputSchema.JSON = json.RawMessage(`{"type":"object","properties":{}}`)
438 out.ToolConfig = &struct {
439 Tools []bedrockToolSpec `json:"tools"`
440 }{Tools: []bedrockToolSpec{stub}}
441 }
442
443 if caching {
444 // Tag the last user message with a cachePoint. This extends the
445 // cached prefix to cover the full conversation history up to the
446 // current turn, so the next turn reads that history cheaply.
447 bedrockTagLastUserCache(out.Messages)
448 }
449
450 return out, nil
451}
452
453// bedrockTagLastUserCache appends a cachePoint block to the last user
454// message in the Bedrock message list. It is the Bedrock equivalent of
455// Anthropic's cache_control:{type:"ephemeral"} on the last user message.
456func bedrockTagLastUserCache(msgs []bedrockMessage) {
457 for i := len(msgs) - 1; i >= 0; i-- {
458 if msgs[i].Role == "user" {
459 msgs[i].Content = append(msgs[i].Content, bedrockCachePoint)
460 return
461 }
462 }
463}
464
465// resolveBedrockInferenceProfileID maps a bare foundation-model ID to
466// its region-matched cross-region inference-profile ID.
467//
468// Several newer Bedrock models (Anthropic Claude 4.x, DeepSeek, etc.)
469// cannot be invoked with on-demand throughput by their plain
470// foundation-model ID; Bedrock returns HTTP 400 demanding "the ID or
471// ARN of an inference profile that contains this model". The profile
472// ID is the same model ID with a geographic prefix (us/eu/apac/...).
473//
474// We only rewrite IDs that (a) lack an existing geo prefix and (b)
475// belong to a model family that requires a profile. IDs that already
476// carry a prefix (e.g. "eu.anthropic...", "global.anthropic...") or
477// fully-qualified ARNs are returned unchanged, so explicit user
478// choices and custom application inference profiles still work.
479func resolveBedrockInferenceProfileID(modelID, region string) string {
480 if modelID == "" {
481 return modelID
482 }
483 // ARNs are already inference-profile references; leave untouched.
484 if strings.HasPrefix(modelID, "arn:") {
485 return modelID
486 }
487 // Already geo-prefixed (us. / eu. / apac. / ap. / us-gov. / global.)?
488 if bedrockHasGeoPrefix(modelID) {
489 return modelID
490 }
491 if !bedrockRequiresInferenceProfile(modelID) {
492 return modelID
493 }
494 prefix := bedrockGeoPrefixForRegion(region)
495 if prefix == "" {
496 return modelID
497 }
498 return prefix + "." + modelID
499}
500
501// bedrockGeoPrefixes are the cross-region inference-profile geo
502// prefixes Bedrock uses. A model ID that starts with one of these
503// (followed by a dot) is already a profile reference.
504var bedrockGeoPrefixes = []string{"us-gov", "us", "eu", "apac", "ap", "global", "au"}
505
506func bedrockHasGeoPrefix(modelID string) bool {
507 for _, p := range bedrockGeoPrefixes {
508 if strings.HasPrefix(modelID, p+".") {
509 return true
510 }
511 }
512 return false
513}
514
515// bedrockRequiresInferenceProfile reports whether a bare
516// foundation-model ID is one of the families AWS only exposes through
517// a cross-region inference profile for on-demand throughput.
518func bedrockRequiresInferenceProfile(modelID string) bool {
519 switch {
520 case strings.HasPrefix(modelID, "anthropic.claude-"):
521 return true
522 case strings.HasPrefix(modelID, "deepseek."):
523 return true
524 default:
525 return false
526 }
527}
528
529// bedrockGeoPrefixForRegion maps an AWS region to the geo prefix used
530// by its cross-region inference profiles. Returns "" when the region
531// has no known mapping, in which case the model ID is left unchanged.
532func bedrockGeoPrefixForRegion(region string) string {
533 switch {
534 case region == "":
535 return "us"
536 case strings.HasPrefix(region, "us-gov-"):
537 return "us-gov"
538 case strings.HasPrefix(region, "us-"):
539 return "us"
540 case strings.HasPrefix(region, "eu-"):
541 return "eu"
542 case strings.HasPrefix(region, "ap-"):
543 return "apac"
544 case strings.HasPrefix(region, "ca-"):
545 return "us"
546 case strings.HasPrefix(region, "sa-"):
547 return "us"
548 default:
549 return "us"
550 }
551}
552
553func (c *bedrockClient) Stream(ctx context.Context, req Request) (<-chan Event, error) {
554 wire, err := c.buildRequest(req)
555 if err != nil {
556 return nil, err
557 }
558 body, err := json.Marshal(wire)
559 if err != nil {
560 return nil, err
561 }
562 modelID := resolveBedrockInferenceProfileID(req.Model, c.region)
563 url := c.baseURL + "/model/" + modelID + "/converse-stream"
564 httpReq, err := http.NewRequestWithContext(ctx, "POST", url, bytes.NewReader(body))
565 if err != nil {
566 return nil, err
567 }
568 httpReq.Header.Set("content-type", "application/json")
569 httpReq.Header.Set("accept", "application/vnd.amazon.eventstream")
570 if c.bearerToken != "" {
571 httpReq.Header.Set("authorization", "Bearer "+c.bearerToken)
572 } else if c.sigv4 != nil {
573 if err := signSigV4(httpReq, body, "bedrock", c.region, c.sigv4, time.Now().UTC()); err != nil {
574 return nil, fmt.Errorf("bedrock: sigv4 sign: %w", err)
575 }
576 } else {
577 return nil, fmt.Errorf("bedrock: no auth configured")
578 }
579 resp, err := c.http.Do(httpReq)
580 if err != nil {
581 return nil, fmt.Errorf("bedrock: %w", err)
582 }
583 if resp.StatusCode != http.StatusOK {
584 b, _ := io.ReadAll(resp.Body)
585 resp.Body.Close()
586 msg := strings.TrimSpace(string(b))
587 // A 403 on the bearer route is almost always a region mismatch:
588 // short-term Bedrock API keys are scoped to the region of the
589 // console session that minted them, but zot defaults to
590 // us-east-1. Surface the resolved region and the fix so the user
591 // is not left guessing why a freshly-copied key is "invalid".
592 if resp.StatusCode == http.StatusForbidden && c.bearerToken != "" {
593 return nil, fmt.Errorf(
594 "bedrock: http 403 (region=%s): %s\nhint: Bedrock API keys are region-scoped. If your key was created in another region, set AWS_REGION (e.g. AWS_REGION=eu-central-1) or pass --base-url https://bedrock-runtime.<region>.amazonaws.com",
595 c.region, msg)
596 }
597 return nil, fmt.Errorf("bedrock: http %d: %s", resp.StatusCode, msg)
598 }
599 out := make(chan Event, 16)
600 go c.runStream(ctx, resp, req, out)
601 return out, nil
602}
603
604func (c *bedrockClient) runStream(ctx context.Context, resp *http.Response, req Request, out chan<- Event) {
605 defer close(out)
606 defer resp.Body.Close()
607
608 out <- EventStart{Provider: "amazon-bedrock", Model: req.Model}
609
610 // State accumulated across deltas.
611 contentBlocks := map[int]*bedrockBlockState{}
612 stop := StopEnd
613 finalMsg := Message{Role: RoleAssistant, Time: time.Now()}
614 var usage Usage
615
616 for {
617 if ctx.Err() != nil {
618 stop = StopAborted
619 break
620 }
621 evt, err := readEventStreamMessage(resp.Body)
622 if err != nil {
623 if err == io.EOF {
624 break
625 }
626 out <- EventDone{Stop: StopError, Err: err, Message: finalMsg}
627 return
628 }
629 eventType := evt.headerString(":event-type")
630 if eventType == "" {
631 eventType = bedrockEventTypeFromPayload(evt.payload)
632 }
633 messageType := evt.headerString(":message-type")
634 if messageType == "exception" {
635 out <- EventDone{Stop: StopError, Err: fmt.Errorf("bedrock exception (%s): %s", evt.headerString(":exception-type"), string(evt.payload)), Message: finalMsg}
636 return
637 }
638 switch eventType {
639 case "messageStart":
640 // nothing to do; role is always "assistant"
641 case "contentBlockStart":
642 var d struct {
643 ContentBlockIndex int `json:"contentBlockIndex"`
644 Start struct {
645 ToolUse *struct {
646 ToolUseID string `json:"toolUseId"`
647 Name string `json:"name"`
648 } `json:"toolUse"`
649 } `json:"start"`
650 }
651 if err := unmarshalBedrockEventPayload(evt.payload, "contentBlockStart", &d); err != nil {
652 continue
653 }
654 st := &bedrockBlockState{}
655 contentBlocks[d.ContentBlockIndex] = st
656 if d.Start.ToolUse != nil {
657 st.isToolUse = true
658 st.toolID = d.Start.ToolUse.ToolUseID
659 st.toolName = d.Start.ToolUse.Name
660 out <- EventToolStart{ID: st.toolID, Name: st.toolName}
661 }
662 case "contentBlockDelta":
663 var d struct {
664 ContentBlockIndex int `json:"contentBlockIndex"`
665 Delta struct {
666 Text string `json:"text"`
667 ToolUse *struct {
668 Input string `json:"input"`
669 } `json:"toolUse"`
670 } `json:"delta"`
671 }
672 if err := unmarshalBedrockEventPayload(evt.payload, "contentBlockDelta", &d); err != nil {
673 continue
674 }
675 st := contentBlocks[d.ContentBlockIndex]
676 if st == nil {
677 st = &bedrockBlockState{}
678 contentBlocks[d.ContentBlockIndex] = st
679 }
680 if d.Delta.Text != "" {
681 st.text.WriteString(d.Delta.Text)
682 out <- EventTextDelta{Delta: d.Delta.Text}
683 }
684 if d.Delta.ToolUse != nil && d.Delta.ToolUse.Input != "" {
685 st.toolArgs.WriteString(d.Delta.ToolUse.Input)
686 out <- EventToolArgs{ID: st.toolID, Delta: d.Delta.ToolUse.Input}
687 }
688 case "contentBlockStop":
689 var d struct {
690 ContentBlockIndex int `json:"contentBlockIndex"`
691 }
692 if err := unmarshalBedrockEventPayload(evt.payload, "contentBlockStop", &d); err != nil {
693 continue
694 }
695 st := contentBlocks[d.ContentBlockIndex]
696 if st == nil {
697 continue
698 }
699 if st.isToolUse {
700 args := st.toolArgs.String()
701 if args == "" {
702 args = "{}"
703 }
704 finalMsg.Content = append(finalMsg.Content, ToolCallBlock{
705 ID: st.toolID, Name: st.toolName, Arguments: json.RawMessage(args),
706 })
707 out <- EventToolEnd{ID: st.toolID}
708 } else if st.text.Len() > 0 {
709 finalMsg.Content = append(finalMsg.Content, TextBlock{Text: st.text.String()})
710 }
711 case "messageStop":
712 var d struct {
713 StopReason string `json:"stopReason"`
714 }
715 _ = unmarshalBedrockEventPayload(evt.payload, "messageStop", &d)
716 switch d.StopReason {
717 case "tool_use":
718 stop = StopToolUse
719 case "end_turn":
720 stop = StopEnd
721 case "max_tokens":
722 stop = StopLength
723 default:
724 stop = StopEnd
725 }
726 case "metadata":
727 var d struct {
728 Usage struct {
729 InputTokens int `json:"inputTokens"`
730 OutputTokens int `json:"outputTokens"`
731 CacheReadInputTokens int `json:"cacheReadInputTokens"`
732 CacheWriteInputTokens int `json:"cacheWriteInputTokens"`
733 } `json:"usage"`
734 }
735 if err := unmarshalBedrockEventPayload(evt.payload, "metadata", &d); err == nil {
736 usage.InputTokens = d.Usage.InputTokens
737 usage.OutputTokens = d.Usage.OutputTokens
738 usage.CacheReadTokens = d.Usage.CacheReadInputTokens
739 usage.CacheWriteTokens = d.Usage.CacheWriteInputTokens
740 if m, err := FindModel("amazon-bedrock", req.Model); err == nil {
741 usage.CostUSD = ComputeCost(m, usage)
742 }
743 out <- EventUsage{Usage: usage}
744 }
745 }
746 }
747 out <- EventDone{Stop: stop, Message: finalMsg}
748}
749
750type bedrockBlockState struct {
751 isToolUse bool
752 toolID string
753 toolName string
754 toolArgs strings.Builder
755 text strings.Builder
756}
757
758func bedrockEventTypeFromPayload(payload []byte) string {
759 var outer map[string]json.RawMessage
760 if err := json.Unmarshal(payload, &outer); err != nil {
761 return ""
762 }
763 for _, name := range []string{"messageStart", "contentBlockStart", "contentBlockDelta", "contentBlockStop", "messageStop", "metadata"} {
764 if _, ok := outer[name]; ok {
765 return name
766 }
767 }
768 return ""
769}
770
771func unmarshalBedrockEventPayload(payload []byte, eventType string, dst any) error {
772 var outer map[string]json.RawMessage
773 if err := json.Unmarshal(payload, &outer); err == nil {
774 if wrapped, ok := outer[eventType]; ok {
775 return json.Unmarshal(wrapped, dst)
776 }
777 }
778 if err := json.Unmarshal(payload, dst); err != nil {
779 return fmt.Errorf("bedrock: parse %s payload: %w", eventType, err)
780 }
781 return nil
782}
783
784// ---- event-stream binary framing parser ----
785
786type eventStreamMessage struct {
787 headers map[string]string
788 payload []byte
789}
790
791func (m *eventStreamMessage) headerString(name string) string { return m.headers[name] }
792
793// readEventStreamMessage reads one event-stream message from r. Returns
794// io.EOF when the stream is finished cleanly.
795func readEventStreamMessage(r io.Reader) (*eventStreamMessage, error) {
796 var prelude [8]byte
797 if _, err := io.ReadFull(r, prelude[:]); err != nil {
798 return nil, err
799 }
800 totalLen := binary.BigEndian.Uint32(prelude[0:4])
801 headersLen := binary.BigEndian.Uint32(prelude[4:8])
802 if totalLen < 16 || headersLen > totalLen-16 {
803 return nil, fmt.Errorf("bedrock: malformed event-stream prelude (total=%d headers=%d)", totalLen, headersLen)
804 }
805 // Skip prelude CRC (4 bytes); we don't validate.
806 var preludeCRC [4]byte
807 if _, err := io.ReadFull(r, preludeCRC[:]); err != nil {
808 return nil, err
809 }
810 headersBuf := make([]byte, headersLen)
811 if _, err := io.ReadFull(r, headersBuf); err != nil {
812 return nil, err
813 }
814 payloadLen := int(totalLen) - 16 - int(headersLen)
815 payload := make([]byte, payloadLen)
816 if _, err := io.ReadFull(r, payload); err != nil {
817 return nil, err
818 }
819 var trailer [4]byte
820 if _, err := io.ReadFull(r, trailer[:]); err != nil {
821 return nil, err
822 }
823 // Parse headers.
824 hdrs := map[string]string{}
825 i := 0
826 for i < len(headersBuf) {
827 if i+1 > len(headersBuf) {
828 break
829 }
830 nameLen := int(headersBuf[i])
831 i++
832 if i+nameLen > len(headersBuf) {
833 return nil, fmt.Errorf("bedrock: header name overflow")
834 }
835 name := string(headersBuf[i : i+nameLen])
836 i += nameLen
837 if i+1 > len(headersBuf) {
838 return nil, fmt.Errorf("bedrock: header value-type missing")
839 }
840 valueType := headersBuf[i]
841 i++
842 // Type 7 = string with 2-byte length. Type 6 = byte-array similar.
843 // Other types (bool/int/long/timestamp/uuid) we treat as opaque
844 // strings since bedrock's event-type headers are always type 7.
845 switch valueType {
846 case 7, 6:
847 if i+2 > len(headersBuf) {
848 return nil, fmt.Errorf("bedrock: header value length missing")
849 }
850 vLen := int(binary.BigEndian.Uint16(headersBuf[i : i+2]))
851 i += 2
852 if i+vLen > len(headersBuf) {
853 return nil, fmt.Errorf("bedrock: header value overflow")
854 }
855 hdrs[name] = string(headersBuf[i : i+vLen])
856 i += vLen
857 case 0: // bool true
858 hdrs[name] = "true"
859 case 1: // bool false
860 hdrs[name] = "false"
861 case 2: // byte
862 if i+1 > len(headersBuf) {
863 return nil, fmt.Errorf("bedrock: header byte missing")
864 }
865 i++
866 case 3: // int16
867 if i+2 > len(headersBuf) {
868 return nil, fmt.Errorf("bedrock: header int16 missing")
869 }
870 i += 2
871 case 4: // int32
872 if i+4 > len(headersBuf) {
873 return nil, fmt.Errorf("bedrock: header int32 missing")
874 }
875 i += 4
876 case 5: // int64
877 if i+8 > len(headersBuf) {
878 return nil, fmt.Errorf("bedrock: header int64 missing")
879 }
880 i += 8
881 case 8: // timestamp (int64)
882 if i+8 > len(headersBuf) {
883 return nil, fmt.Errorf("bedrock: header timestamp missing")
884 }
885 i += 8
886 case 9: // uuid (16 bytes)
887 if i+16 > len(headersBuf) {
888 return nil, fmt.Errorf("bedrock: header uuid missing")
889 }
890 i += 16
891 default:
892 return nil, fmt.Errorf("bedrock: unknown header value-type %d", valueType)
893 }
894 }
895 return &eventStreamMessage{headers: hdrs, payload: payload}, nil
896}
897
898// ---- SigV4 signing ----
899//
900// AWS Signature Version 4. Implemented from scratch so we don't pull in
901// aws-sdk-go-v2 (large transitive dep tree just for one signer). The
902// algorithm is documented at:
903// https://docs.aws.amazon.com/general/latest/gr/sigv4-signed-request-examples.html
904//
905// Steps:
906//
907// 1. Build the canonical request: method + path + canonical query +
908// canonical headers + signed headers + payload-hash.
909// 2. Build the string-to-sign: algorithm + amz-date + credential scope +
910// sha256(canonical request).
911// 3. Derive the signing key by chaining HMAC-SHA256:
912// kSecret -> kDate -> kRegion -> kService -> kSigning
913// 4. signature = HMAC-SHA256(kSigning, stringToSign)
914// 5. Authorization header: AWS4-HMAC-SHA256 Credential=..., SignedHeaders=..., Signature=...
915//
916// Payload hash is sha256 of the JSON body. For Bedrock's streaming
917// endpoint this works because the request body is fully buffered before
918// the call — we don't need the streaming-payload variant.
919
920func sigHMAC(key, msg []byte) []byte {
921 m := hmac.New(sha256.New, key)
922 m.Write(msg)
923 return m.Sum(nil)
924}
925
926func sigSHA256Hex(b []byte) string {
927 h := sha256.Sum256(b)
928 return hex.EncodeToString(h[:])
929}
930
931func signSigV4(req *http.Request, payload []byte, service, region string, creds *bedrockSigV4Creds, now time.Time) error {
932 amzDate := now.Format("20060102T150405Z")
933 shortDate := now.Format("20060102")
934 credentialScope := shortDate + "/" + region + "/" + service + "/aws4_request"
935
936 // Required headers.
937 req.Header.Set("host", req.URL.Host)
938 req.Header.Set("x-amz-date", amzDate)
939 if creds.sessionToken != "" {
940 req.Header.Set("x-amz-security-token", creds.sessionToken)
941 }
942 payloadHash := sigSHA256Hex(payload)
943 req.Header.Set("x-amz-content-sha256", payloadHash)
944
945 // Canonical headers: lowercase name, trimmed value, sorted by name,
946 // each on its own line ending with \n.
947 var names []string
948 values := map[string]string{}
949 for k, vs := range req.Header {
950 lower := strings.ToLower(k)
951 // Skip authorization (we're computing it) and any
952 // connection-specific headers.
953 if lower == "authorization" {
954 continue
955 }
956 names = append(names, lower)
957 values[lower] = strings.Join(vs, ",")
958 }
959 // host is special — it's not in req.Header by default; set above.
960 sort.Strings(names)
961 var canonHdrs strings.Builder
962 for _, n := range names {
963 canonHdrs.WriteString(n)
964 canonHdrs.WriteByte(':')
965 canonHdrs.WriteString(strings.TrimSpace(values[n]))
966 canonHdrs.WriteByte('\n')
967 }
968 signedHeaders := strings.Join(names, ";")
969
970 // Canonical query string: keys sorted by name, each key and value
971 // URI-escaped (RFC 3986).
972 q := req.URL.Query()
973 var qKeys []string
974 for k := range q {
975 qKeys = append(qKeys, k)
976 }
977 sort.Strings(qKeys)
978 var canonQuery strings.Builder
979 for i, k := range qKeys {
980 if i > 0 {
981 canonQuery.WriteByte('&')
982 }
983 canonQuery.WriteString(url.QueryEscape(k))
984 canonQuery.WriteByte('=')
985 canonQuery.WriteString(url.QueryEscape(q.Get(k)))
986 }
987
988 // Canonical request.
989 canonReq := req.Method + "\n" +
990 canonicalSigV4Path(req.URL.Path) + "\n" +
991 canonQuery.String() + "\n" +
992 canonHdrs.String() + "\n" +
993 signedHeaders + "\n" +
994 payloadHash
995
996 // String to sign.
997 stringToSign := "AWS4-HMAC-SHA256\n" + amzDate + "\n" + credentialScope + "\n" + sigSHA256Hex([]byte(canonReq))
998
999 // Signing key.
1000 kDate := sigHMAC([]byte("AWS4"+creds.secretAccessKey), []byte(shortDate))
1001 kRegion := sigHMAC(kDate, []byte(region))
1002 kService := sigHMAC(kRegion, []byte(service))
1003 kSigning := sigHMAC(kService, []byte("aws4_request"))
1004
1005 // Signature.
1006 signature := hex.EncodeToString(sigHMAC(kSigning, []byte(stringToSign)))
1007
1008 // Authorization header.
1009 auth := "AWS4-HMAC-SHA256 " +
1010 "Credential=" + creds.accessKeyID + "/" + credentialScope + ", " +
1011 "SignedHeaders=" + signedHeaders + ", " +
1012 "Signature=" + signature
1013 req.Header.Set("Authorization", auth)
1014 return nil
1015}
1016
1017// canonicalSigV4Path returns the path with each segment URI-encoded
1018// according to SigV4 rules (everything except unreserved chars
1019// A-Za-z0-9-_.~ and / between segments). Bedrock model IDs contain
1020// dots, colons, and slashes; the colons must be percent-encoded.
1021func canonicalSigV4Path(p string) string {
1022 if p == "" {
1023 return "/"
1024 }
1025 segments := strings.Split(p, "/")
1026 for i, s := range segments {
1027 segments[i] = sigV4EncodeSegment(s)
1028 }
1029 return strings.Join(segments, "/")
1030}
1031
1032func sigV4EncodeSegment(s string) string {
1033 var b strings.Builder
1034 for i := 0; i < len(s); i++ {
1035 c := s[i]
1036 // Unreserved per RFC 3986.
1037 if (c >= 'A' && c <= 'Z') || (c >= 'a' && c <= 'z') || (c >= '0' && c <= '9') ||
1038 c == '-' || c == '_' || c == '.' || c == '~' {
1039 b.WriteByte(c)
1040 continue
1041 }
1042 fmt.Fprintf(&b, "%%%02X", c)
1043 }
1044 return b.String()
1045}