Repository navigation
plugin(providers): AWS Bedrock LLM provider — Claude, Titan, Llama, Mistral via Amazon Bedrock #9010
Description
Summary
Allow AutoBot to route LLM calls through AWS Bedrock. Enables access to Claude models for AWS-committed customers and provides access to AWS-exclusive models (Titan, Nova) and open-source models (Llama, Mistral) via Bedrock's hosted inference.
What Users Want
bedrockprovider type in LLM gateway config- Model support: Claude 3.x/4.x, Llama 3.x, Mistral, Amazon Titan/Nova
- AWS credential auth (access key + secret, or IAM role via instance profile)
- Region selection
- Cross-region inference profiles for better availability
- Cost tracking via Bedrock pricing (per-token, varies by model)
AutoBot Fit
New bedrock.py in autobot-backend/llm_shared/providers/ using boto3 (bedrock-runtime client). Streaming via invoke_model_with_response_stream. AWS credentials stored in secrets system. Plugin installable from Marketplace.
Acceptance Criteria
- Bedrock provider selectable in LLM gateway config
- Access key and IAM role auth both supported
- Claude, Llama, and Mistral models working with streaming
- Region and cross-region inference profile configurable
- Cost tracking with Bedrock on-demand pricing
Activity
feat(llm): AWS Bedrock provider with Claude, Llama, Mistral, Titan, N…
feat(llm): register BedrockProvider in provider registry (#9010)
mrveiss commentedon Jun 17, 2026
Closed — verify-and-close. The AWS Bedrock provider is already implemented and registered in
Dev_new_gui; all acceptance criteria are met by existing code:BedrockProviderregistered in the provider registry (llm_shared/provider_registry.py:733-769), credential-gated; surfaced dynamically via/api/llm-providers/fallback-status(frontendLLMProvidersView.vuelists registered providers — no hardcoded list).types.py:42 BEDROCK = "bedrock".AWS_ACCESS_KEY_ID/SECRETand IAM-role fallback wrapped in try/except so missing creds never crash startup;boto3import is lazy insideBedrockProvider.__init__).BEDROCK_MODELS(5 Claude, 7 Llama, 4 Mistral, + Titan/Nova) viainvoke_model_with_response_stream(llm_shared/providers/bedrock.py).settings["region"]/AWS_DEFAULT_REGION+ inference profiles.bedrock.py.No code change required. Discoveries: none.