neuralmind

NeuralMind vs. Aider repo-map

What Aider’s repo-map is

Aider builds a concise, tree-sitter-derived map of your repository — a ranked list of the most relevant symbols and their signatures — and injects it into the prompt every turn. It’s a clever, zero-embedding, purely static approach.

How NeuralMind differs

Dimension	Aider repo-map	NeuralMind
Technique	Static tree-sitter + PageRank over symbol graph	Knowledge graph + vector embeddings (ChromaDB)
Semantic retrieval	No — syntactic signal only	Yes — embedding similarity over all nodes
Output	Ranked signatures injected every turn	On-demand progressive layers (wakeup / query / skeleton)
Host	Aider CLI only	Any MCP agent + CLI + copy-paste to any LLM
Tool-output compression	None	Read/Bash/Grep PostToolUse hooks
Learns from usage	No	Yes — a Hebbian synapse layer learns associations automatically from queries, edits, and tool calls (with decay)
Languages	Whatever tree-sitter supports	Same (via graphify)
Install methods	pip install aider-chat + LLM API key	pip / pipx / uv / Docker / source — no API key required for the local index

When to pick which

• Pick Aider repo-map if you use Aider and want a zero-dependency, deterministic map.
• Pick NeuralMind if you want semantic (not just syntactic) retrieval, want it outside Aider, or want the consumption-side compression. Aider’s map answers “what symbols exist”; NeuralMind’s query answers “which 800 tokens best explain X”.

They compose: you can feed NeuralMind’s wakeup output into an Aider session as additional context.

---

← Back to comparison index

This site is open source. Improve this page.