Skip to content

The cross-harness, self-improving memory layer for AI agents. Join our community: https://discord.com/invite/zdNChS2fBu

1.7 KB
Raw Blame
# Smoke config: mocked models, so no provider key and no spend. claude-tools answers with a
# tool call reading a credentials file, which the post_call guardrail records.
model_list:
  - model_name: claude-mock
    litellm_params:
      model: anthropic/claude-sonnet-4-5
      api_key: sk-dummy
      mock_response: "ok"
  - model_name: claude-tools
    litellm_params:
      model: anthropic/claude-sonnet-4-5
      api_key: sk-dummy
      # The Anthropic route mocks only when mock_response is set; the tool call rides along.
      mock_response: "Reading your AWS credentials."
      mock_tool_calls:
        - id: toolu_smoke_read
          type: function
          function:
            name: Read
            arguments: '{"file_path": "/home/dev/.aws/credentials"}'
  - model_name: gpt-mock
    litellm_params:
      model: openai/gpt-4o
      api_key: sk-dummy
      mock_response: "ok"

  # A real provider-shaped reply: scripts/smoke/upstream answers every call with a tool_use.
  - model_name: claude-upstream
    litellm_params:
      model: anthropic/claude-sonnet-4-5
      api_key: sk-dummy
      api_base: http://127.0.0.1:4020

guardrails:
  - guardrail_name: beacon-gateway-request
    litellm_params:
      guardrail: generic_guardrail_api
      mode: pre_call
      api_base: http://127.0.0.1:4019
      default_on: true
      unreachable_fallback: fail_open
      extra_headers: [x-claude-code-session-id, session_id]
  - guardrail_name: beacon-gateway-response
    litellm_params:
      guardrail: generic_guardrail_api
      mode: post_call
      api_base: http://127.0.0.1:4019
      default_on: true
      unreachable_fallback: fail_open
      extra_headers: [x-claude-code-session-id, session_id]
      streaming_buffer_until_moderated: false