[Feature Request]: Add AWS Bedrock support

[wgong]

Do you need to file a feature request?

• I have searched the existing feature request and this feature request is not already filed.
• I believe this is a legitimate feature request, not just a question or bug.

Feature Request Description

We are interested in evaluating DeepCode at work, where AWS Bedrock is used,
it would be nice to support this new LLM provider

I can test this feature when implemented

Additional Context

No response

[lihuiyang1024]

I rechecked this against the current AWS interfaces before implementing it. Amazon Bedrock now exposes OpenAI-compatible Chat Completions and Responses endpoints on bedrock-runtime, in addition to the native Converse API:

• https://docs.aws.amazon.com/bedrock/latest/userguide/models-api-compatibility.html
• https://docs.aws.amazon.com/bedrock/latest/userguide/inference-chat-completions-mantle.html
• https://docs.aws.amazon.com/bedrock/latest/userguide/conversation-inference.html

That leaves an important provider-contract choice:

1. OpenAI-compatible first: add a Bedrock connection template using a region-specific /openai/v1 base URL and an Amazon Bedrock API key. This reuses DeepCode’s existing adapter and adds no boto3 dependency.
2. Native AWS first: add a Converse/ConverseStream adapter using the standard AWS credential chain (profiles, IAM roles, SSO/temporary credentials). This better matches enterprise AWS use, but adds a new backend, dependency, model discovery path, and tool-call conversion surface.

My preference for a focused first PR is option 1, with explicit documentation that SigV4/IAM-role authentication is not yet covered. If IAM roles/SSO are required for this issue’s intended workplace use, I’ll design the native adapter instead. Which contract do you want DeepCode to support first?

[wgong]

My teammate already implemented AWS bedrock integration, he can definitely give you PR feedback Thanks, Wen

…

On Sat, Sep 5, 2026, 6:24 AM YANGYU ***@***.***> wrote: *lihuiyang1024* left a comment (HKUDS/DeepCode#101) <#101 (comment)> I rechecked this against the current AWS interfaces before implementing it. Amazon Bedrock now exposes OpenAI-compatible Chat Completions and Responses endpoints on bedrock-runtime, in addition to the native Converse API: - https://docs.aws.amazon.com/bedrock/latest/userguide/models-api-compatibility.html - https://docs.aws.amazon.com/bedrock/latest/userguide/inference-chat-completions-mantle.html - https://docs.aws.amazon.com/bedrock/latest/userguide/conversation-inference.html That leaves an important provider-contract choice: 1. *OpenAI-compatible first:* add a Bedrock connection template using a region-specific /openai/v1 base URL and an Amazon Bedrock API key. This reuses DeepCode’s existing adapter and adds no boto3 dependency. 2. *Native AWS first:* add a Converse/ConverseStream adapter using the standard AWS credential chain (profiles, IAM roles, SSO/temporary credentials). This better matches enterprise AWS use, but adds a new backend, dependency, model discovery path, and tool-call conversion surface. My preference for a focused first PR is option 1, with explicit documentation that SigV4/IAM-role authentication is not yet covered. If IAM roles/SSO are required for this issue’s intended workplace use, I’ll design the native adapter instead. Which contract do you want DeepCode to support first? — Reply to this email directly, view it on GitHub <#101?email_source=notifications&email_token=AACQRSHIVAHWMT6FGSBPUL35NPSW5A5CNFSNUABFM5UWIORPF5TWS5BNNB2WEL2JONZXKZKDN5WW2ZLOOQXTKNJVGEYTKNBRGA3KM4TFMFZW63VGMF2XI2DPOKSWK5TFNZ2KYZTPN52GK4S7MNWGSY3L#issuecomment-5551154106>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AACQRSGZOYSMC4YFIP5D6AD5NPSW5AVCNFSNUABFKJSXA33TNF2G64TZHM4TQMZSGQ3DKNRYHNEXG43VMU5TGNZUHE2TKNBQGQ2KC5QC> . You are receiving this because you authored the thread.Message ID: ***@***.***>

[Zongwei9888]

Thanks for checking the current AWS surface before writing code — that is the right call. Let's go with option 1, OpenAI-compatible first.

Some framing for the PR:

• Today a Bedrock connection already works through the custom template (OpenAI-compatible backend with a user-supplied base URL and key), so the PR is really a template plus catalog work: a ProviderSpec for Bedrock with requires_api_base=True (the registry has no region placeholder, so the user supplies the region-specific /openai/v1 URL), the AWS_BEARER_TOKEN_BEDROCK key convention, catalog entries for the model IDs Bedrock actually exposes (anthropic.claude-…, the cross-region us./eu. prefixes), and a short guide section.
• Please state in the docs that SigV4 / IAM roles / SSO are not covered by this first step; the native Converse adapter can be a separate discussion once the template has landed and people have tried it.
• @wgong — if your teammate can review the template PR against your workplace setup, that would be very helpful.