Turn one long Claude conversation into a navigable mind map using a DSPy Recursive Language Model (RLM).

python -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt
# Deno (required by DSPy's RLM sandbox):
curl -fsSL https://deno.land/install.sh | sh
cp .env.example .env   # then fill in GEMINI_API_KEY (default), or another provider's key

threadmap build examples/sample_conversation.md --out out/ --trace

Outputs land in out/: mindmap.html, mindmap.md, mindmap.json, and (with --trace) trace.txt.

Defaults: gemini/gemini-2.5-pro (root) + gemini/gemini-2.5-flash (sub). DSPy routes through LiteLLM, so any LiteLLM-supported model works. Override via env:

THREADMAP_ROOT_MODEL=anthropic/claude-opus-4-5 \
THREADMAP_SUB_MODEL=anthropic/claude-haiku-4-5 \
threadmap build mychat.md --out out/

The CLI checks for the env var matching your root model's provider (GEMINI_API_KEY, ANTHROPIC_API_KEY, OPENAI_API_KEY, etc.) and errors clearly if it's missing.