Sonde
A free, open-source local code-graph engine that lets AI coding agents query symbol relationships and code structure directly, instead of guessing from repeated grep-style searches.
🔗 Visit Sonde
Description
AI coding agents often 'explore' a codebase by running search after search, burning tool calls and context just to figure out how one function relates to another. Sonde builds an actual structural map of your code up front, so an agent can query it directly instead of guessing.
Sonde indexes TypeScript, Python, and Swift codebases into a symbol-level graph stored in SQLite, and exposes it to AI agents through three MCP tools: find_symbols, query_graph, and get_impact_radius. It requires zero manual setup — indexing happens locally — and it reports drift if the graph falls out of sync with the actual code. The project's own benchmark claims 8x fewer tool calls and 3x less context usage than typical agentic search loops, with perfect recall on structural questions. It's Apache 2.0 licensed, free, with no hosted service.
💬 Our review
The short version: Sonde targets a real inefficiency in how AI coding agents currently explore code — repeated, expensive search — and its claimed 8x reduction in tool calls is a meaningful efficiency gain if it holds up in your codebase.
Sourcegraph offers far more mature, large-scale code intelligence and search, but it's a heavier, often paid product built for big organizations, not a lightweight local MCP server. Universal ctags and similar symbol-indexing tools have existed for decades and can build a similar structural index, but they weren't designed with AI-agent MCP access as the primary interface, so wiring them into an agent workflow takes more manual work than Sonde's out-of-the-box MCP tools. The realistic alternative most agents use today is exactly what Sonde is trying to replace: iterative grep and file-reading loops, which work but cost tool calls and context. For TypeScript, Python, or Swift projects where you're running agents heavily, Sonde is a low-cost (free, local, zero-setup) experiment worth running to see if it actually cuts down agent exploration overhead.
💰 Pricing
📊 Global score
🤖 AI-enriched data
Open source sous licence Apache 2.0, aucun coût, pas de service hébergé
Pros
8x moins d'appels d'outils selon le benchmark du projet
Trois outils MCP prêts à l'emploi (find_symbols, query_graph, get_impact_radius)
Indexation locale sans configuration
Rapport de dérive si le graphe se désynchronise du code
Cons
Limité à TypeScript, Python et Swift
Pas de service hébergé, tout en local
Projet jeune, moins mature que Sourcegraph