Skip to content
Discussion options

You must be logged in to vote

Hi @rumleyo,

According to the official docs/api.md, Colibri runs 100% locally (pure C inference engine). The model itself cannot access your filesystem through plain chat — it needs an agent layer that provides tools like read_file and write_to_file.

Why it thinks it's in the cloud

The LLM weights (e.g., Qwen) keep their pre-training belief that they run on Alibaba Cloud. Running locally doesn't change that unless you give it filesystem tools via Tool Calling / Function Calling.

Official solution: Use an agentic VS Code extension

  1. Start the OpenAI-compatible server:
    cd c
    COLI_MODEL=/path/to/your/model ./coli serve --host 127.0.0.1 --port 8000 --model-id glm-5.2-colibri
  2. In VS Code, instal…

Replies: 1 comment 1 reply

Comment options

You must be logged in to vote
1 reply
@rumleyo
Comment options

Answer selected by rumleyo
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Category
Q&A
Labels
None yet
2 participants