William Easton
William Easton
If there is agreement on this I'd be happy to offer a PR for the GenAI LLM and the Orchestrator
Actually looking at this again, I think this flow is probably broken? The original generation produces a plan result, and the original text completion is unaware of the desired response...
I'm not sure you even need instructor as a fallback option, it's just using the genai sdk and passing a response model schema that it has parsed, there's nothing instructor...
Switching to instructor could basically eliminate all tool preparation, message handling, response formatting, and tool call handling. I think you want to keep control of the tool calls, but switching...
Thank you! That helps! How would you feel about an implementation where we require tool calls and expose two add't tools: 1. Report Completion w/ response model 2. Report Failure...
actually i think a server is just failing to start and not giving me any info on why
Looks like maybe NPX servers arent supported via config dicts yet? ``` git: command: "npx" args: ["-y", "@cyanheads/git-mcp-server"] description: "Perform git-related actions" ```
If you use pedantic models for the parameters the descriptions will be in the schema. ``` class FunctionParams(BaseModel): arg1 = Field(description="cool description") def function(args: FunctionParams): pass ``` Also normally the...
In my server I have a second version of most tools which allows bulk actions, so I have one version which allows creating a knowledge base which takes primitives as...