Problem Statement
Currently, open router is used and output parsing is relied solely on prompting, which is fragile.
Proposed Solution
Use BAML by wrapping openrouter endpoint: https://docs.boundaryml.com/ref/llm-client-providers/openai-generic
It would be better to allow to replace the LLM endpoint so that the user can use local LLMs like ollama.
FYI, from claude agent sdk side, you can use structured output feature for the same purpose: https://platform.claude.com/docs/en/agent-sdk/structured-outputs
Alternatives Considered
No response
Feature Area
Policy engine / evaluation
Related Harness
None
Use Case
No response
Additional Context
No response
Problem Statement
Currently, open router is used and output parsing is relied solely on prompting, which is fragile.
Proposed Solution
Use BAML by wrapping openrouter endpoint: https://docs.boundaryml.com/ref/llm-client-providers/openai-generic
It would be better to allow to replace the LLM endpoint so that the user can use local LLMs like ollama.
FYI, from claude agent sdk side, you can use structured output feature for the same purpose: https://platform.claude.com/docs/en/agent-sdk/structured-outputs
Alternatives Considered
No response
Feature Area
Policy engine / evaluation
Related Harness
None
Use Case
No response
Additional Context
No response