v0.1.2 on PyPI · works with Codex · Claude Code · Cursor · Copilot→

Graph-native context for AI coding agents

Map your repo once. Vinemap injects the exact files your agent needs — before the first tool call, with hooks for Claude, Cursor, Codex, and more. 100% on your machine.

[MCP][PRE-INJECT][LOCAL]
Install Free →View on GitHubv0.1.2

100% local — nothing leaves your machine · free for individual developers · no account, no API keys

vinemap — mcp
Live retrieval41ms
$ graph_retrieve('auth flow')
→auth.tsmiddlewareapi/
pack: 842 tokens · injected to agent

MCP ready for

ClaudeCodexCursorGeminiCopilotOpenCode
Get started

Install in under a minute.

One global install via PyPI or our GitHub installers, then run on any project. Python 3.9+ is the only requirement.

<60sSetup time
2Commands
0Accounts needed
1
Global install~15 sec

Install once

Pure Python, zero dependencies — works offline, never breaks on a transitive pin.

bash
# PyPI — pure Python, zero dependencies
$ pip install vinemap
# or one-line installer from GitHub
$ curl -fsSL https://raw.githubusercontent.com/rohit0x62/vinemap/main/installers/install.sh | bash
2
Per project~45 sec

Run in your project

Run vinemap index . in your project directory to build the graph, then vinemap connect cursor (or claude, gemini, codex) to wire up your AI tool automatically. Or run vinemap quickstart . for a guided wizard.

terminal
$ vinemap index .
indexed 1,204 files — 8,913 symbols, 3,412 import edges in 1.9s
$ vinemap connect cursor
wrote .cursor/mcp.json
✓ graph tools live — graph_retrieve · graph_read · graph_neighbors
→ reload your agent and ask away
ClaudeCodexCursorGeminiCopilotOpenCode
🔒 100% local — nothing leaves your machine.

Works inside any editor

MCP runs over stdio — no plugin required.

VS CodeJetBrainsNeovimSublimeZedTerminal

…and any other editor with a terminal — Emacs, Helix, Fleet, you name it.

Under the hood

A dual graph of your project.

Vinemap builds two layers — a structural map of your code and a live memory of your session — and uses both to deliver precise context.

Code Map

Your project’s DNA

A complete graph of your codebase — every file, function, and class, and how they connect. Built in seconds, updated incrementally.

  • Files, functions, classes with line ranges
  • Import and call edges across files
  • Tree-sitter (optional) or regex parsers for 12 languages
  • Monorepo cluster awareness (package.json, go.mod, pyproject)
Context Packer

The intelligence layer

Compresses graph results into a compact structured summary — not raw file dumps. Your agent gets more understanding in fewer tokens.

  • Full signatures with params and returns
  • BM25 over symbols + structural expansion
  • Optional local embeddings as a fourth signal
  • Hard token budgets — the pack always fits
Session Memory

Smarter every turn

Tracks what has been read, edited, and decided. Follow-up questions route straight to previously relevant files.

  • Touched files weighted higher next turn
  • vinemap decide records WHY across sessions
  • Dashboard tracks token savings per session
  • Guardrails curb blind grep/read loops
Platform

Built for real agent workflows.

Beyond MCP tools — pre-injection hooks, live indexing, guardrails, and a reproducible eval harness. See measured retrieval quality on our benchmarks page.

Pre-injection

Context before tool call #1

Agent-specific hooks for Claude Code, Cursor rules, Codex, Gemini, Copilot, and OpenCode — packs land in the prompt before grep or Read.

Parsing

Tree-sitter + regex fallback

Python AST by default; optional tree-sitter for TS/JS, Go, Rust, Java, C/C++, C#, Ruby, PHP, Kotlin, Swift. Install with pip install vinemap[treesitter].

Ranking

Graph + BM25 + session boost

Lexical BM25 over symbols, structural expansion along import/call edges, monorepo cluster awareness, and optional local embeddings.

Watch

Live graph on save

vinemap watch re-indexes changed files with debounced sub-second updates — no manual rebuilds after every edit.

Guardrails

Fewer blind reads

Per-turn read budgets, duplicate-read dedup, and grep nudges steer agents toward graph_retrieve instead of exploration loops.

Dashboard

Token savings, visible

vinemap dashboard shows touched files, session decisions, and estimated tokens saved. Add --web for a local UI.

The exploration tax

Stop paying for blind search.

Without Vinemap

  1. You ask a question
  2. Agent calls grep, Bash, Read
  3. Reads 10–20 files to find context
  4. More exploration, more tool calls…
  5. Finally answers your question

Thousands of tokens wasted on exploration, every turn.

With Vinemap

  1. You ask a question
  2. The graph packs the correct context instantly
  3. Agent answers with full context

One pass. Zero exploration overhead. Better answers.

Compare

Why a graph beats a dump, a map, or a cloud.

Repo dumps send everything. Embedding RAG guesses by similarity. Cloud engines need your code on their servers. Vinemap routes exact structural context locally.

Structural graphImports, symbols, and call edges
100% localCode never leaves your machine
Session memoryTouched files + WHY decisions
Pre-injectionContext packed before the first tool call
BM25 + structureNot just embedding similarity
Live re-indexGraph stays fresh on save
Repomix-style

Repo dump

Token firehose

Ships entire files — most of it irrelevant noise.

Aider-style

Repo map

Static snapshot

Shows structure but no live session recall.

Similarity search

Embedding RAG

Guesswork retrieval

Finds lookalikes, not the actual call chain.

Remote index

Cloud engine

Off-device

Your codebase indexed on someone else's servers.

Vinemap

Graph + pre-injection

All four dimensions

Structural graph, session memory, and packed context — 100% on your machine.

Feature matrixVinemap leads on every axis
Supported Missing
Approach
GraphStructural graph
Local100% local
MemorySession memory
InjectPre-injection
Score
Vinemapgraph + pre-injectionBest fit
4/4
Repo dumpsRepomix
—
—
—
1/4
Repo mapsAider
—
—
2/4
Embedding RAGContinue.dev
—
—
—
1/4
Cloud context enginesAugment · Greptile · Cody
—
—
2/4

← Swipe to compare features →

PythonTypeScriptJavaScriptGoJavaRustC / C++C#RubyPHPKotlinSwift
Pricing

Free for developers. Teams when you scale.

Every feature — unlimited project size, diagnosis, audit, health checks — is free for individual use. We'll introduce paid plans later for org-wide Teams features.

Individual
Free

Full engine · no file limits · forever

  • Graph-first context + all MCP tools
  • Pre-injection hooks for 6 agents
  • Crash diagnosis, health checks, audit mode
  • Session memory + WHY decisions
  • vinemap watch live re-index
  • Token dashboard + guardrails
  • 100% local & private
Install free →
Coming soon
Teams
Custom

For engineering orgs · self-hosted or VPC

  • Self-hosted Docker graph server
  • Cross-repo retrieval across services
  • Shared WHY memory with attribution
  • OIDC SSO + audit log + seat management
  • Priority onboarding & direct support
Join the waitlist
Creator

Built by an engineer who's shipped RAG at scale

Vinemap comes from years of building production AI platforms — not a weekend MCP wrapper.

Rohit Jha

Creator of Vinemap

Engineering Lead · AI Platform & Distributed Systems

Bengaluru, India

Connect on LinkedIn

Principal backend engineer with 8+ years building cloud-native, high-scale systems — multi-tenant SaaS platforms serving 3M+ users and 60M+ monthly transactions at 99.9% uptime.

In production I've built RAG pipelines, LLM fine-tuning, and conversational AI that moved retrieval accuracy from 65% to 92%. Vinemap applies that same obsession with context quality — but structural, local, and built for coding agents instead of cloud indexes.

3M+Users on platforms architected
92%Production RAG accuracy (from 65%)
60M+Monthly transactions at scale
8+ yrsBackend & AI platform engineering
Engineering LeadNexxio Technologies

B2B SaaS · RAG knowledge bases · LLM fine-tuning · vector search

Senior Software Engineer (founding backend)GoodWorker

Job platform · microservices · fraud & ranking systems

RAG & LLMsDistributed systemsAWS & serverlessMCP / agent toolingMulti-tenant SaaS
LIVE NOW

Join the Vinemap community

Connect with other developers on GitHub Discussions, share your setup, and get help configuring Vinemap for your codebase.

GitHub Discussions
REAL FEEDBACK

Got feedback or ideas?

We read every message. Open a GitHub issue or email us — custom plans, larger codebases, or setup help.

Open an issue
FAQ

Frequently asked questions

Does my code leave my machine?

No. The graph, session memory, and context packs are stored in .vinemap/ inside your project. The MCP server runs locally over stdio — no code, no file names, no project data is ever sent externally.

Is Vinemap free?

Yes — completely free for individual developers. No file limits, no account, no API keys. All features are included: MCP tools, pre-injection hooks, crash diagnosis, health checks, audit mode, session memory, and the token dashboard. Paid Teams plans for orgs are coming later.

Which AI tools does Vinemap work with?

Claude Code, Codex CLI, Cursor, Gemini CLI, GitHub Copilot, OpenCode, and any other MCP-compatible agent. Run vinemap cursor . or vinemap claude . for full setup with pre-injection hooks, or vinemap connect <agent> for MCP-only wiring.

Do I need to run something every session?

No. Run vinemap index . once, then vinemap connect <agent> or vinemap quickstart . — after that your agent talks to the local server automatically. Optional: vinemap watch . keeps the graph fresh on every save.

What happens when my files change?

Content-hash caching means only touched files are re-parsed — updates are sub-second on most projects. For hands-free freshness, run vinemap watch . in a terminal tab; it debounces saves and re-indexes incrementally.

What languages and parsers are supported?

Python uses full AST parsing. TypeScript, JavaScript, Go, Rust, Java, C/C++, C#, Ruby, PHP, Kotlin, and Swift use regex symbol extraction by default, or precise tree-sitter parsing with pip install vinemap[treesitter].

How is ranking different from embedding RAG?

Vinemap ranks with a structural code graph — imports, symbols, and call edges — plus BM25 over identifiers, monorepo cluster awareness, session memory, and optional local embeddings. It finds the actual call chain, not text lookalikes.

How will Teams pricing work?

Teams (coming soon) adds a self-hosted Docker graph server for orgs: cross-repo retrieval, shared WHY memory with attribution, OIDC SSO, audit logs, and seat management. Individual use stays free — Teams is for engineering organizations that need a shared graph in their VPC.

Can I trust the retrieval quality claims?

Yes — we publish reproducible precision@k benchmarks on vinemap.xyz/benchmarks, exported from the open-source eval harness in engine/eval/. CI fails if scores regress. Run python eval/run_eval.py --all locally to verify.

Give your codebase a brain.

One global install, then run on any project. Free for individual developers.

Install Free →
💬 Ask Vinemap?