Source Code Indexing Comparison

Key Conclusion

The 4 tools in the graph comparison (semantica, trustgraph, mempalace, neural-memory) are NOT good choices for source code indexing.

Among the 14 repos, graphify, code-review-graph, and gitnexus are the three that specialize in source code. graphify parses 23 languages with tree-sitter, keeps a persistent graph, and lists a 71.5× token reduction. code-review-graph covers 19 languages plus Jupyter, stores the index in SQLite, and focuses on blast radius. gitnexus is the zero-server option, with LadybugDB and a browser UI. The other four are built for documents or general memory, not for code indexing.

These tools focus on:

  • semantica: Document-based KG, reasoning, audit trail
  • trustgraph: Multi-model data, enterprise knowledge
  • mempalace: Memory system, Palace structure
  • neural-memory: Graph-based memory, spreading activation

Best Tools for Source Code Indexing

Among the 14 repos, 3 tools specialize in source code:

graphify ⭐ Best

Feature Support
Languages 23 languages (Python, JS, TS, Go, Rust, Java, C++, etc.)
AST Parsing ✅ tree-sitter
Code Indexing ✅ Persistent graph (graph.json)
Code Search ✅ query, path, explain commands
Token Reduction 71.5×
MCP ✅ (MCP server)
Output graph.html (interactive), GRAPH_REPORT.md

How it works:

  1. AST pass - deterministic extraction (classes, functions, imports, call graphs)
  2. Semantic pass - Claude extracts concepts, relationships, design rationale
  3. Output: NetworkX graph + Leiden community detection

code-review-graph

Feature Support
Languages 19 languages + Jupyter
AST Parsing ✅ tree-sitter
Code Indexing ✅ SQLite
Code Search ✅ Blast-radius, impact analysis
Token Reduction 8.2× - 49×
MCP ✅ (22 tools)
Output D3.js visualization

Strengths:

  • Incremental updates (<2s)
  • 100% recall on impact analysis
  • Auto-update on file edit/git commit

gitnexus

Feature Support
Languages Multiple (tree-sitter based)
AST Parsing ✅ tree-sitter
Code Indexing ✅ LadybugDB
Code Search ✅ MCP tools (query, context, impact)
MCP ✅ (16 tools)
Architecture Zero-server (WASM)

Strengths:

  • Runs in browser (WASM)
  • Zero-server architecture
  • Enterprise features (PR review, wiki)

Detailed Comparison

Feature graphify code-review-graph gitnexus semantica trustgraph mempalace neural-memory
Languages 23 19+ Multiple Generic Generic ❌ ❌
AST Parsing ✅ tree-sitter ✅ tree-sitter ✅ tree-sitter ⚠️ basic ❌ ❌ ❌
Code Indexing ✅ ✅ ✅ ❌ ❌ ❌ ❌
Code Search ✅ ✅ ✅ ❌ ❌ ❌ ❌
Token Reduction 71.5× 8.2-49× - - - - -
MCP ✅ ✅ (22) ✅ (16) ❌ ✅ ✅ ✅ (56)
Offline ✅ ✅ ✅ ✅ ✅ ✅ ✅

Recommendations

For Source Code Indexing & Searching

Use Case Recommendation Reason
Codebase understanding graphify 71.5× token reduction, 23 languages, AST parsing
Code review code-review-graph Blast-radius, impact analysis
Code intelligence gitnexus Zero-server, WASM, enterprise
Simple code search code-review-graph Fast, incremental updates
Tool Reason
semantica Document-focused, no AST parsing
trustgraph Enterprise knowledge, not code analysis
mempalace Memory system, no code features
neural-memory Memory library, not focused on code

Quick Decision

What do you need?
│
├─► Index & Search Source Code
│   ├─► Token optimization (71.5×) → graphify
│   ├─► Code review, impact analysis → code-review-graph
│   ├─► Zero-server, WASM → gitnexus
│   ├─► Speed + auto-sync, MIT → codegraph
│   ├─► Most languages, zero-dependency → codebase-memory-mcp
│   └─► Graph + git history + code health + decisions → repowise
│
├─► Knowledge Graph / Semantic Search (documents)
│   ├─► Audit, reasoning → semantica
│   ├─► Multi-model enterprise → trustgraph
│   └─► Memory system → mempalace / neural-memory
│
└─► Vector Search Only
    └─► zvec (high-performance)

Conclusion: To index and search a project's source code, use graphify (best for token reduction), code-review-graph (for review), gitnexus (for enterprise, accepting its noncommercial license), codegraph (fastest, auto-sync, pure MIT), codebase-memory-mcp (most languages, zero-dependency), or repowise (if you need git analytics/code health/decisions and not just a graph).


September 2026 Update — 3 New Tools (repowise, codegraph, codebase-memory-mcp)

These three tools were not in the original comparison table (04/2026) because they appeared or became popular afterward. See each tool's details: Repowise, CodeGraph, Codebase Memory MCP.

Tool Core language Languages parsed MCP tools License Differentiator
repowise Python 26 10 (task-shaped) AGPL-3.0/commercial Graph + git analytics + code health + ADR mining, PR bot
codegraph Rust kernel 20 focused on codegraph_explore MIT Fastest, auto-sync via OS file watcher, self-scales to the machine
codebase-memory-mcp Pure C 162 15 MIT Zero-dependency (single binary), Hybrid LSP, makes no network calls itself

A note on trustworthiness: all three (and gitnexus as well) are very young (created 08/2025–03/2026) yet have reached 6.7k–71.5k GitHub stars — the ratio of watchers/followers to stars is lower than typical for long-established popular repos. This is not evidence of fraud, but you should verify further (commit history, community discussion) and avoid piping curl | bash/curl | sh straight into a shell where possible, before fully trusting the self-published benchmark numbers/badges in a README.

An AST pass extracts classes, functions, imports, and call graphs. A semantic pass asks Claude for concepts, relationships, and design rationale. code-review-graph updates in under 2 seconds and states 100% recall on impact analysis, with 22 MCP tools and a token cut between 8.2× and 49×. repowise parses 26 languages and adds git analytics, code health, and ADR mining under AGPL-3.0. codegraph is a Rust kernel with an OS file watcher and an MIT license. codebase-memory-mcp is a single C binary that parses 162 languages and makes no network calls itself.

Updated: 2026-04-13, supplemented 2026-09-20

See also: Graph Comparison · Memory Tool Comparison · Graphify