Skip to content

Latest commit

 

History

History
20 lines (11 loc) · 876 Bytes

File metadata and controls

20 lines (11 loc) · 876 Bytes

Model providers

Status: accepted

Supported providers

  • Azure OpenAI through Azure Responses-compatible endpoints.
  • AWS Bedrock through the AWS credential chain and converse-compatible models.

Configuration

The dashboard configures provider, model/deployment, endpoint, region and a secret reference name. It never stores secret values. A hello action makes a bounded request and creates a local OTEL span; its UI response currently contains the provider response only.

Policy invocation

The evaluation phase renders the versioned policy prompt with bounded evidence and validates the returned response schema. Source-changing policy actions are not implemented.

Planned work

Expose hello latency/model metadata/trace ID, configure token and reasoning limits, and enforce policy output safety constraints before any future source-changing action.