bedrock support

[darinkishore]

What's happening?

there is currently no maintained bedrock support. we can't see exactly how to add in bedrock support, and use it with Claude models.

the docs the URL points to on anthropic OAuth are very out of date :(

Where does this happen?

None

Impact on your workflow

None

Environment (if applicable)

No response

Additional context

No response

[github-actions]

Thank you for raising this. Here's my assessment:

Current Status

Big-AGI has partial AWS Bedrock support via the Anthropic vendor's "API Host" field (visible in Advanced settings), but there are significant limitations:

What Works (Limited)

• The UI references Bedrock at src/modules/llms/vendors/anthropic/AnthropicServiceSetup.tsx:93 with a placeholder for a docs/config-aws-bedrock.md that does not exist
• You can set a custom API host (e.g., bedrock-runtime.us-east-1.amazonaws.com)

What Doesn't Work

1. Authentication: Bedrock uses AWS Signature v4 (IAM credentials), not Anthropic API keys. Big-AGI only supports X-API-Key header auth (see anthropic.access.ts:133)
2. Different API endpoints: Bedrock uses /model/{model-id}/invoke vs Anthropic's /v1/messages
3. Beta features: Bedrock doesn't support Anthropic's beta headers (prompt caching, extended thinking, etc.) which Big-AGI sends automatically
4. Model listing: The Anthropic /v1/models endpoint doesn't exist on Bedrock

Recommended Solutions

Option 1: Use a Bedrock Proxy (Recommended for now)

Deploy a proxy that translates Anthropic API calls to Bedrock:

• bedrock-claude-proxy - Converts Anthropic API format to Bedrock
• bedrock-access-gateway - OpenAI-compatible gateway

Then point Big-AGI's Anthropic "API Host" to your proxy.

Option 2: Use LiteLLM Proxy

LiteLLM can proxy Bedrock through an OpenAI-compatible endpoint. Configure it as an OpenAI-compatible vendor in Big-AGI.

Option 3: Native Bedrock Integration (Feature Request)

This would require implementing:

• AWS Signature v4 signing
• Bedrock-specific endpoint routing
• Handling of Bedrock inference profiles
• Selective disabling of unsupported beta features

This is a substantial effort (~2-3 weeks of dev work).

Next Steps

Could you clarify your use case?

1. Are you looking for a quick workaround (proxy approach)?
2. Would native Bedrock support be valuable for your team/organization?
3. Are you using specific Bedrock features (inference profiles, VPC endpoints)?

This will help prioritize the right approach.

---

References:

• Claude on Amazon Bedrock - Anthropic Docs
• Anthropic Claude Messages API - AWS Docs
• bedrock-claude-proxy