feat: add support for anthropic to AWS bedrock for Anthropic translation

[secustor]

Description

This PR implement adds a translator from the Anthropic API to AWS Bedrock for Anthropic.

Related Issues/PRs (if applicable)

Closes #1371

Special notes for reviewers (if applicable)

Has been tested on our side together with #1394 successfully E2E using Claude Code

Using following setup

export ANTHROPIC_BASE_URL=http://localhost:8080/anthropic 
export ANTHROPIC_AUTH_TOKEN=dummy
export ANTHROPIC_MODEL=eu.anthropic.claude-haiku-4-5-20251001-v1:0
export CLAUDE_CODE_MAX_OUTPUT_TOKENS=4096
export MAX_THINKING_TOKENS=1024

claude

The implementation is more or less a copy of the Anthropic to GCP implementation, so it would make sense to include it in the planned refactor with the Anthropic to Anthropic translator

[codecov-commenter]

Codecov Report

❌ Patch coverage is 86.66667% with 4 lines in your changes missing coverage. Please review.
✅ Project coverage is 83.67%. Comparing base (f547bad) to head (c6c6aad).
⚠️ Report is 83 commits behind head on main.

Files with missing lines	Patch %	Lines
internal/extproc/messages_processor.go	33.33%	2 Missing ⚠️
...ernal/extproc/translator/anthropic_awsanthropic.go	92.59%	1 Missing and 1 partial ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1418      +/-   ##
==========================================
- Coverage   83.67%   83.67%   -0.01%     
==========================================
  Files         138      139       +1     
  Lines       12268    12297      +29     
==========================================
+ Hits        10265    10289      +24     
- Misses       1395     1400       +5     
  Partials      608      608              

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[seluard]

Latest refactor changes tested already at our setup 🧪 ✅

[mathetake]

Thanks, this is exciting and thank you for testing locally. can you add tests in here after you resolve my comments?

ai-gateway/tests/extproc/testupstream_test.go

Lines 821 to 887 in 1f8ca94

	{
	name: "gcp-anthropicai - /anthropic/v1/messages",
	backend: "gcp-anthropicai",
	path: "/anthropic/v1/messages",
	method: http.MethodPost,
	requestBody: `{"model":"claude-3-sonnet","max_tokens":100,"messages":[{"role":"user","content":[{"type":"text","text":"Hello, just a simple test message."}]}],"stream":false}`,
	expRequestBody: `{"anthropic_version":"vertex-2023-10-16","max_tokens":100,"messages":[{"content":[{"text":"Hello, just a simple test message.","type":"text"}],"role":"user"}],"stream":false}`,
	expHost: "gcp-region-aiplatform.googleapis.com",
	expPath: "/v1/projects/gcp-project-name/locations/gcp-region/publishers/anthropic/models/claude-3-sonnet:rawPredict",
	expRequestHeaders: map[string]string{"Authorization": "Bearer " + fakeGCPAuthToken},
	responseStatus: strconv.Itoa(http.StatusOK),
	responseBody: `{"id":"msg_123","type":"message","role":"assistant","stop_reason": "end_turn", "content":[{"type":"text","text":"Hello from native Anthropic API!"}],"usage":{"input_tokens":8,"output_tokens":15}}`,
	expStatus: http.StatusOK,
	expResponseBody: `{"id":"msg_123","type":"message","role":"assistant","stop_reason": "end_turn", "content":[{"type":"text","text":"Hello from native Anthropic API!"}],"usage":{"input_tokens":8,"output_tokens":15}}`,
	},
	{
	name: "gcp-anthropicai - /anthropic/v1/messages - streaming",
	backend: "gcp-anthropicai",
	path: "/anthropic/v1/messages",
	method: http.MethodPost,
	responseType: "sse",
	requestBody: `{"model":"claude-3-sonnet","max_tokens":100,"messages":[{"role":"user","content":[{"type":"text","text":"Tell me a short joke"}]}],"stream":true}`,
	expRequestBody: `{"anthropic_version":"vertex-2023-10-16","max_tokens":100,"messages":[{"content":[{"text":"Tell me a short joke","type":"text"}],"role":"user"}],"stream":true}`,
	expHost: "gcp-region-aiplatform.googleapis.com",
	expPath: "/v1/projects/gcp-project-name/locations/gcp-region/publishers/anthropic/models/claude-3-sonnet:streamRawPredict",
	expRequestHeaders: map[string]string{"Authorization": "Bearer " + fakeGCPAuthToken},
	responseStatus: strconv.Itoa(http.StatusOK),
	responseBody: `event: message_start
	data: {"type": "message_start", "message": {"id": "msg_789", "usage": {"input_tokens": 8}}}
	event: content_block_start
	data: {"type": "content_block_start", "index": 0, "content_block": {"type": "text", "text": ""}}
	event: content_block_delta
	data: {"type": "content_block_delta", "index": 0, "delta": {"type": "text_delta", "text": "Why don't scientists trust atoms? Because they make up everything!"}}
	event: content_block_stop
	data: {"type": "content_block_stop", "index": 0}
	event: message_delta
	data: {"type": "message_delta", "delta": {"stop_reason": "end_turn"}, "usage": {"output_tokens": 15}}
	event: message_stop
	data: {"type": "message_stop"}
	`,
	expStatus: http.StatusOK,
	expResponseBody: `event: message_start
	data: {"type": "message_start", "message": {"id": "msg_789", "usage": {"input_tokens": 8}}}
	event: content_block_start
	data: {"type": "content_block_start", "index": 0, "content_block": {"type": "text", "text": ""}}
	event: content_block_delta
	data: {"type": "content_block_delta", "index": 0, "delta": {"type": "text_delta", "text": "Why don't scientists trust atoms? Because they make up everything!"}}
	event: content_block_stop
	data: {"type": "content_block_stop", "index": 0}
	event: message_delta
	data: {"type": "message_delta", "delta": {"stop_reason": "end_turn"}, "usage": {"output_tokens": 15}}
	event: message_stop
	data: {"type": "message_stop"}
	`,
	},

[secustor]

yet not validated e2e, but should work based on the tests

[secustor]

My implementation relies on the ( current ) behaviour that when setting the same multiple headers with the same keys the last one wins.

I have added this test to ensure that behaviour in the future isn't broken by accident

[mathetake]

sorry can you remove this? it's irrelevant here in this PR. As I noted we do not want to mess up the commit trees with a conflated change. You wouldn't want to end up finding completely unrelated commit in a file that you are debugging

[seluard]

yet not validated e2e, but should work based on the tests

e2e failed in our side with the latest changes. The duplicated content-length header seems to break the AWS signature, let us work to fix in our side before you take a look.

[mathetake]

you would need to dedup content-length here @secustor @seluard

[mathetake]

kindly ping @seluard @secustor -- we are planning to cut v0.4 very soon, but I would love to see this land. Would it be possible to do the follow here to reach the finish line, or can I just do the clean up and push the changes here for you? Thank you!

[secustor]

@mathetake Added the requested changes, but we couldn't test e2e so far because of unrelated constraints.

Looking forward to v0.4.0 then.

[mathetake]

Thank you for your contribution @secustor (and @seluard too)!

I cleaned up and added the e2e tests too, and verified it's working in my AWS env. good to go!