Memtrace provides structural memory for AI coding agents, transforming codebases into a live, bi-temporal knowledge graph. It enables multiple agents to query shared code context in milliseconds without LLM calls, facilitating collaborative development and safe refactoring by offering full causal awareness and version history. This addresses the need for efficient, scalable, and cost-effective code intelligence for AI-driven development workflows.

How It Works

Memtrace leverages Rust and Tree-sitter to parse codebases into a bi-temporal graph, indexing symbols and relationships locally and deterministically. Unlike LLM-dependent systems, it avoids inference costs and latency by directly parsing ASTs. The system stores full version history for each code element, enabling "time-travel" queries and precise impact analysis. It also maps cross-service API topologies, providing a comprehensive structural understanding of complex architectures.

Quick Start & Requirements

• Install: npm install -g memtrace
• Prerequisites: Node.js ≥ 18, Git.
• Resources: Minimum 4 cores, 8 GB RAM, 5 GB disk; Recommended 8+ cores, 16-32 GB RAM, 10-20 GB disk for large repos.
• Docs: memtrace.io, Docs, npm, Discord

Highlighted Details

• Indexes 1,500 files in ~1.5s ($0 cost), ~1,200x faster than alternatives.
• Achieves 0.07 ms exact symbol query latency, 128x lower than GitNexus.
• Memory footprint is 41x tighter than ChromaDB (26 MB vs 1,060 MB).
• Supports 20+ languages and framework-specific scanners (e.g., Terraform, GitHub Actions, Django).
• Features LeanCTX Native (v0.3.57+) for advanced compression and adaptive learning.

Maintenance & Community

Memtrace is currently in private beta, with access rolled out in batches to maintain a tight feedback loop. Users can join the waitlist at memtrace.io. The project is built by Syncable. Community support is available via Discord.

Licensing & Compatibility

The project operates under a proprietary EULA, free for individual developers during beta and post-general availability. The core indexer and database (MemDB) are closed-source. It integrates with various AI coding agents (Claude Code, Cursor, Codex) via the Model Context Protocol (MCP).

Limitations & Caveats

The core indexing engine and database are closed-source. The project is currently in a private beta phase, requiring users to join a waitlist for access.