memtrace
Free面向编码代理的结构化内存层，通过 AST 构建双时态知识图谱，为 Claude、Cursor 等 MCP 客户端提供低延迟代码理解与检索能力。
FreeFree tier
About memtrace
Memtrace turns your codebase into a live knowledge graph that AI coding agents can query in milliseconds — every function, class, call edge, and version, across every session, without re-reading files or breaking things they can't see. It uses Rust + Tree-sitter to index up to 50k files in under 90 seconds with zero LLM calls, fully local. Supports bi-temporal time-travel queries, replay-aware refactoring, and shared memory for fleets of coding agents. Available for Cursor, Claude Code, Codex, Hermes, VS Code, and Windsurf via MCP.
Key Features
Zero LLM calls and no API costs
Bi-temporal time-travel queries to see code history
Index 50k files in under 90 seconds
Rust + Tree-sitter, supports 20+ languages
MCP-native integration with Cursor, Claude Code, etc.
Replay-aware refactoring with full causal awareness
LeanCTX Native compressed reads for efficiency
Fully local execution, code never leaves machine
Shared structural memory for multiple agents prevents merge conflicts
Pros & Cons
Pros
- Zero LLM calls, eliminating API costs
- Extremely fast indexing (50k files in <90s) and queries (0.07ms latency)
- High accuracy (96.6% exact symbol query, 81.6% graph callers recall)
- Low memory footprint (26 MB RSS)
- Supports 20+ programming languages with framework-aware scanners
- Fully local, preserving code privacy
- Bi-temporal history enables time-travel queries
- Shared memory across agents prevents merge conflicts and stale context
Cons
- Currently in private beta, limited access to new users
- Requires MCP-compatible client (e.g., Claude Code, Cursor, Windsurf) for integration
- Documentation and community resources are still evolving
- Indexing may have overhead for extremely large or complex repos beyond benchmarks
Best For
Running a fleet of coding agents on the same repo without merge conflictsReplaying refactors with full causal awareness and blast radius analysisAccelerating code review with PR context and dependency trackingUnderstanding codebase dependencies and call graphs for large projectsEnhancing AI coding agents with low-latency structural understanding
FAQ
What is Memtrace?
Memtrace is an open-source structural memory layer that turns your codebase into a live knowledge graph for AI coding agents. It enables millisecond queries of functions, classes, and call edges without re-reading files.
How fast is indexing?
Memtrace can index up to 50,000 files in under 90 seconds. For 1,500 files, it takes about 1.5 seconds, roughly 1,200 times faster than alternatives like Mem0.
Is my code sent to the cloud?
No, Memtrace runs fully locally. Your code never leaves your machine, as stated in the documentation.
What languages and tools are supported?
Memtrace supports 20+ languages via Tree-sitter, and includes framework-aware scanners for Vapor, Lapis, Kong, GitHub Actions, Terraform, and RLS policies. It integrates via MCP with Cursor, Claude Code, Codex, Hermes, VS Code, and Windsurf.
What is the pricing?
Memtrace is open source and free to use. It is currently in private beta, with access granted in batches.