← all repositories
giancarloerra/SocratiCode

A search engine that actually reads your code before answering

SocratiCode indexes your entire codebase so AI assistants can search semantically, trace dependencies, and understand blast radius instead of grepping blindly.

SocratiCode
Velocity · 7d
+28
★ / day
Trend
steady
star history

What it does SocratiCode is a local, Docker-backed context engine that plugs into any MCP host (Claude Code, Cursor, VS Code, etc.). It indexes your codebase with AST-aware chunking, builds polyglot dependency graphs across 18 languages, and serves hybrid semantic + BM25 search plus symbol-level impact analysis. Your AI stops guessing where to look and starts knowing.

The interesting bit The heavy lifting happens upfront: resumable batched indexing checkpoints progress, a file watcher keeps everything current, and the engine pre-computes call-flow and blast radius. The README claims this lets smaller models tackle architectural tasks that normally require top-tier reasoning — a pragmatic bet that compute spent indexing beats compute wasted on bad context.

Key highlights

  • Zero-config local deployment via Docker; Qdrant and Ollama spin up automatically on first run
  • Hybrid search fuses vector embeddings with BM25 (RRF), plus cross-project and branch-aware queries
  • Interactive HTML graph explorer for visual dependency and circular-dependency navigation
  • Ingests “context artifacts” — database schemas, API specs, infra configs, architecture docs — alongside code
  • Benchmarked on 2.45M LOC (VS Code): 61% less context, 84% fewer tool calls, 37× faster than grep-based exploration with Claude Opus 4.6
  • AGPL-3.0 licensed; cloud-hosted beta with SSO/air-gapped options exists but the core stays free

Caveats

  • First indexing is not instant: ~5 minutes for setup plus download, and up to 10 minutes for 3M+ LOC on an M4 MacBook Pro
  • macOS/Windows Docker performance on large codebases is flagged as a potential concern (truncated in sources)
  • npx caching means updates require manual cache clearing or accepting slower startup with @latest

Verdict Worth a look if you’re tired of watching your AI assistant flail through grep results on a large or polyglot codebase. Skip it if your project is small enough that ripgrep and a good CLAUDE.md already suffice.

heatdrop uses Google Analytics to see which pages get read — nothing else. Your call. How we handle data.