grove-explore-base

The local code-locator delegate for grove’s delegated exploration mode (grove init --as mcp-llm). grove’s inner Rust explorer drives a short tool-calling loop against this model over an OpenAI-compatible endpoint and returns validated file:line citations.

Untrained base. Off-the-shelf Qwen3.5-4B (Alibaba/Qwen), self-converted to GGUF and quantized — not a grove fine-tune.

Pull

ollama pull bonigopalan/grove-explore-base:q4_k_m   # default, 2.78 GB
ollama pull bonigopalan/grove-explore-base:q8_0     # baseline, 4.6 GB

Point grove at it — .grove/explore.json

{
  "provider": "ollama",
  "base_url": "http://localhost:11434/v1",
  "model": "bonigopalan/grove-explore-base:q4_k_m",
  "steering": "strict"
}

Serving notes

• Thinking is ON (default on ollama ≥ 0.32) — required; thinking off yields empty tool-calls.
• Baked-in params: temperature 0, num_ctx 24576.

Requires ollama ≥ 0.32 (earlier engines reject this qwen35 GGUF). Apache-2.0, inherited from upstream Qwen3.5-4B.