memtrace

What it does

Three things, every release.

🧭   Run a fleet of coding agents on the same repo without merge hell. Each agent reads the same call graph, sees the same blast radius, inherits the same temporal history. No collisions. No stale context.

🔁   Replay any refactor with full causal awareness. Agents see exactly what depends on what, and what changed when. No more "I refactored a function and 14 tests broke that nobody saw."

⚡   Index a 50k-file repo in under 90 seconds. Rust + Tree-sitter, $0 in API costs, 20+ languages plus framework-aware scanners (Vapor, Lapis, Kong, GitHub Actions, Terraform, RLS policies, …), fully local. Your code never leaves your machine.

🆕   LeanCTX Native — compressed reads, smart trees, and a value ledger. Four new compression modes on get_source_window, single-call directory maps, real-time token-savings dashboard, and an opt-in adaptive learner that beats the static table by ~14%. Full breakdown: docs/leanctx-native.md. Available in v0.3.57+.

https://github.com/user-attachments/assets/e7d6a1e9-c912-4e65-a421-bd0256dffa5a

Related MCP server: remembrallmcp

Numbers

Reproducible benchmark suite: benchmarks/. Same machine, same corpora, same adapter contract. Ground truth from Python's ast and pyright LSP — never from any tool's own index. No system gets a home-field advantage in the dataset.

Detailed breakdowns: BENCHMARKS-v0.3.22.md · BENCHMARKS-v0.3.29.md · Code reviewer benchmark

GitHub Star Growth

Get access

Memtrace is in private beta. We're rolling out access in batches to keep the feedback loop tight — every cohort lands in a Discord channel where we ship fixes from real bug reports inside a week.

→ Join the waitlist at memtrace.io.

Already have access? npm install -g memtrace and you're indexing in 90 seconds. Full setup below.

🔒 Privacy. Memtrace runs entirely on your machine. Source code never leaves it. The only network traffic is license validation, aggregate node/edge counts, and opt-out crash telemetry — no source, no file paths, no symbol names. Full breakdown: PRIVACY.md, TELEMETRY.md. Disable telemetry with MEMTRACE_TELEMETRY=off.

Why Memtrace exists

Good code-intelligence tools already exist. GitNexus and CodeGrapherContext build AST-based graphs that work for "what's in my repo right now."

Memtrace is a bi-temporal episodic structural knowledge graph. It builds on the same AST foundation and adds two dimensions:

• Temporal memory — every symbol carries its full version history. Six scoring algorithms (impact, novelty, recency, directional, compound, overview) let agents ask different temporal questions: "what changed?", "what's unexpected?", "what'll break?".

• Cross-service API topology — Memtrace maps HTTP call graphs between repositories, detecting which services call which endpoints across your architecture.

On top of that, the structural layer is comprehensive:

The agent doesn't just search your code. It remembers it.

Memtrace vs. general memory systems (Mem0, Graphiti)

Mem0 and Graphiti are strong conversational memory engines designed for tracking entity knowledge (e.g. User -> Likes -> Apples). They excel at that. For code intelligence specifically, the tradeoff is that they rely on LLM inference to build their graphs — which adds cost and time when processing thousands of source files.

Graphiti processes data through add_episode(), which triggers multiple LLM calls per episode — entity extraction, relationship resolution, deduplication. At ~50 episodes/minute (source), ingesting 1,500 code files takes 1–2 hours.

Mem0 processes data through client.add(), which queues async LLM extraction and conflict resolution per memory item (source). Bulk ingestion with infer=True (default) means every file passes through an LLM pipeline. Throughput is bounded by your LLM provider's rate limits.

Both accumulate $10–50+ in API costs for large codebases because every relationship is inferred rather than parsed.

Memtrace takes a different approach: it indexes 1,500 files in 1.2–1.8 seconds for $0.00 — no LLM calls, no API costs, no rate limits. Native Tree-sitter AST parsers resolve deterministic symbol references (CALLS, IMPLEMENTS, IMPORTS) locally. The tradeoff is that Memtrace is purpose-built for code — it doesn't handle conversational entity memory the way Mem0 and Graphiti do.

25+ MCP tools

Memtrace exposes a full structural toolkit via the Model Context Protocol.

Search & Discovery

• find_code — hybrid BM25 + semantic + RRF

• find_symbol — exact / fuzzy with Levenshtein

Relationships

• analyze_relationships — callers, callees, hierarchy, imports

• get_symbol_context — 360° view in one call

Impact Analysis

• get_impact — blast radius with risk rating

• detect_changes — diff-to-symbols scope mapping

Code Quality

• find_dead_code — zero-caller detection

• find_most_complex_functions — complexity hotspots

• calculate_cyclomatic_complexity

• get_repository_stats

Temporal Analysis

• get_evolution — 6 scoring modes

• get_timeline — full version history

• detect_changes — diff-based scope

Graph Algorithms

• find_bridge_symbols — betweenness centrality

• find_central_symbols — PageRank / degree

• list_communities — Louvain modules

• list_processes / get_process_flow

API Topology

• get_api_topology — cross-repo HTTP graph

• find_api_endpoints

• find_api_calls

Indexing & Watch

• index_directory — parse, resolve, embed

• watch_directory — live incremental

• execute_cypher — direct graph queries

17 agent skills

Memtrace ships skills/guidance that teach agents how to use the graph. They fire automatically based on what you ask — no prompt engineering required.

Plus 8 workflow skills that chain multiple tools with decision logic: memtrace-first, codebase-exploration, change-impact-analysis, incident-investigation, refactoring-guide, continuous-memory, episode-replay, and session-continuity.

Temporal Engine

Six scoring algorithms for different temporal questions:

Uses Structural Significance Budgeting to surface the minimum set of changes covering ≥80% of total significance.

Compatibility

Skills are workflow prompts that teach the agent how to chain tools. Kiro does not use SKILL.md, so Memtrace writes equivalent auto steering files instead.

Setup

Claude Code + Claude Desktop

npm install -g memtrace

Handles everything — binary, 17 skills, MCP server, plugin, marketplace. One command, both editors.

For manual setup:

claude plugin marketplace add https://github.com/syncable-dev/memtrace-public.git
claude plugin install memtrace-skills@memtrace --scope user
claude mcp add memtrace -- memtrace mcp

Cursor

npm install -g memtrace handles everything automatically. Cursor v2.4+ reads the same SKILL.md format as Claude.

For project-local install (skills travel with your repo):

npx memtrace-skills install --only cursor --local

Codex, Windsurf, VS Code, Hermes, OpenCode, and Kiro

The installer also writes skills/guidance and MCP configuration for the newer agent surfaces:

Install only selected integrations:

npx memtrace-skills install --only codex,windsurf,vscode,hermes,opencode,kiro

Install project-local config where supported:

npx memtrace-skills install --only codex,vscode,opencode,kiro --local

Other MCP clients

For Cline, Roo Code, or any client that only needs MCP tools, add this server manually:

{
  "mcpServers": {
    "memtrace": {
      "command": "memtrace",
      "args": ["mcp"],
      "env": {}
    }
  }
}

Uninstall

memtrace uninstall      # removes skills, MCP server, plugin, settings
npm uninstall -g memtrace

Already ran npm uninstall first? The cleanup script is at ~/.memtrace/uninstall.js:

node ~/.memtrace/uninstall.js

Install troubleshooting

npm install -g memtrace ships a small main package + a platform-specific binary (one of @memtrace/darwin-arm64, @memtrace/linux-x64, @memtrace/win32-x64). If memtrace start ever says "Could not find binary for your platform":

# Re-run install, asking npm to keep optional deps
npm install -g memtrace --include=optional

# Or refresh from latest
memtrace install         # built-in self-update
npm install -g memtrace@latest --force

# Or install the platform binary directly (Apple Silicon shown — swap for your platform)
npm install -g @memtrace/darwin-arm64

This typically only happens on machines where npm is configured to skip optional dependencies (corporate npmrc, certain CI caches).

Languages

Programming: Rust · Go · TypeScript · JavaScript · Python · Java · C · C++ · C# · Swift · Kotlin · Ruby · PHP · Dart · Scala · Perl · Lua — full AST: functions, classes, types, calls, complexity.

Infrastructure & config: YAML · HCL / Terraform · JSON · TOML · SQL (including PostgreSQL CREATE POLICY for RLS, with cross-language edges from policies to Drizzle / Prisma / TS schema symbols).

Framework-aware scanners on top of the AST layer:

• Backend HTTP: Express · NestJS · Encore · Fastify · Vapor · Hummingbird · FastAPI · Flask · Django · Gin · Chi · Echo · Actix · Lapis · Kong · OpenResty · Rails routes

• Frontend / client: RTK Query · TanStack Query · SWR · URLSession · AsyncHTTPClient · axios · fetch · SwiftUI views

• CI / infra: GitHub Actions workflows (jobs, steps, needs: edges) · Terraform variables / modules / data sources · Helm charts · K8s manifests

• Package & dependency graphs: package.json scripts + deps · Cargo.toml deps · pyproject.toml (best-effort)

• Database: PostgreSQL RLS policies + triggers + functions, with heuristic edges to ORM schema

Requirements

Memtrace runs locally — first index is CPU/RAM intensive, subsequent queries and incremental indexing are much lighter.

Telemetry

Since v0.3.17 Memtrace ships with opt-out telemetry that helps us catch crashes, regressions, and performance issues before someone files an issue.

• Collected: app-start events, indexing/embedding durations, panic reports, WARN/ERROR log lines from Memtrace's own crates.

• NOT collected: source code, file contents, symbol names, embeddings, repository names or paths, branch names, commit data.

• Sanitisation: every payload is run through a sanitiser that strips home-dir paths, token-shaped strings, and email addresses before it touches disk.

Disable with one env var:

MEMTRACE_TELEMETRY=off memtrace start                    # per-run
export MEMTRACE_TELEMETRY=off                             # permanent (~/.zshrc, ~/.bashrc)

Or in your editor's MCP config: "env": { "MEMTRACE_TELEMETRY": "off" }.

Full breakdown — including the on-disk queue layout, where data is stored on the receiving end, and how to inspect what would have shipped — is in TELEMETRY.md.

License & ownership

Proprietary EULA. Free to use during private beta and after general availability for individual developers. Indexer + database (MemDB) are closed-source.

Benchmark suite under MIT in benchmarks/ — fully reproducible, no proprietary code required to run them.