<p align="center"> <img src=".github/m1nd-logo.svg" alt="m1nd" width="400" /> </p> <h3 align="center">Stop letting AI grep your codebase. Give it a neuro-symbolic Connectome.</h3> <p align="center"> m1nd simulates counterfactual node removals, epidemic defect spread, and standing wave harmonics to find bugs LLMs can't see.<br/> <em>(Local execution. 84% fewer tokens. Your code as a 4D living graph with Hebbian plasticity.)</em> </p> <p align="center"> <strong>39 missing bugs found in one session · 89% hypothesis accuracy · Saves 84% in LLM context costs</strong> </p> <p align="center"> <a href="https://crates.io/crates/m1nd-core"><img src="https://img.shields.io/crates/v/m1nd-core.svg" alt="crates.io" /></a> <a href="https://github.com/maxkle1nz/m1nd/actions"><img src="https://github.com/maxkle1nz/m1nd/actions/workflows/ci.yml/badge.svg" alt="CI" /></a> <a href="LICENSE"><img src="https://img.shields.io/badge/license-MIT-blue.svg" alt="License" /></a> <a href="https://docs.rs/m1nd-core"><img src="https://img.shields.io/docsrs/m1nd-core" alt="docs.rs" /></a> </p> <p align="center"> <a href="#why-this-changes-everything">Why This Changes Everything</a> · <a href="#quick-start">Quick Start</a> · <a href="#configure-your-agent">Configure Your Agent</a> · <a href="#proven-results">Results</a> · <a href="#the-61-tools">Tools</a> · <a href="https://github.com/maxkle1nz/m1nd/wiki">Wiki</a> · <a href="EXAMPLES.md">Examples</a> </p> <h4 align="center">Works with any MCP client</h4> <p align="center"> <a href="https://claude.ai/download"><img src="https://img.shields.io/badge/Claude_Code-f0ebe3?logo=claude&logoColor=d97706" alt="Claude Code" /></a> <a href="https://cursor.sh"><img src="https://img.shields.io/badge/Cursor-000?logo=cursor&logoColor=fff" alt="Cursor" /></a> <a href="https://codeium.com/windsurf"><img src="https://img.shields.io/badge/Windsurf-0d1117?logo=windsurf&logoColor=3ec9a7" alt="Windsurf" /></a> <a href="https://github.com/features/copilot"><img src="https://img.shields.io/badge/GitHub_Copilot-000?logo=githubcopilot&logoColor=fff" alt="GitHub Copilot" /></a> <a href="https://zed.dev"><img src="https://img.shields.io/badge/Zed-084ccf?logo=zedindustries&logoColor=fff" alt="Zed" /></a> <a href="https://github.com/cline/cline"><img src="https://img.shields.io/badge/Cline-000?logo=cline&logoColor=fff" alt="Cline" /></a> <a href="https://roocode.com"><img src="https://img.shields.io/badge/Roo_Code-6d28d9?logoColor=fff" alt="Roo Code" /></a> <a href="https://github.com/continuedev/continue"><img src="https://img.shields.io/badge/Continue-000?logoColor=fff" alt="Continue" /></a> <a href="https://opencode.ai"><img src="https://img.shields.io/badge/OpenCode-18181b?logoColor=fff" alt="OpenCode" /></a> <a href="https://aws.amazon.com/q/developer"><img src="https://img.shields.io/badge/Amazon_Q-232f3e?logo=amazonaws&logoColor=f90" alt="Amazon Q" /></a> </p>

Architecture & Paradigm Shift

Until now, AI agents have navigated code like humans using a terminal: searching for text (grep), listing files (ls), or relying on static RAG vectors that don't understand causality.

m1nd replaces text search with physics and epidemiology.

When an AI agent uses m1nd's 61 MCP tools across your codebase, it's not searching text. It's injecting a signal into a living graph and watching it propagate across structural, semantic, temporal, and causal dimensions.

84% Fewer Tokens: Stop shoving entire files into the context window. m1nd surgically extracts only the topologically relevant paths, slashing token costs and preventing context exhaustion. Instead of 210 grep operations reading 80,000 lines, m1nd gives the agent the exact graph node in 3 milliseconds using 0 tokens.
Epidemic Bug Detection: m1nd uses the SIR epidemiological model to predict where the next bug will spawn based on the infection rate of connected modules.
Counterfactual Simulation: "What happens if I remove this node?" m1nd simulates the blast radius before the AI even touches the file. It finds synergistic failures that LLMs simply cannot see.
Hebbian Plasticity: The graph learns. When the AI gets a correct answer, the neural pathways forming that connection physically strengthen. The next query is radically faster and smarter.
Context Efficiency: Paradoxically, m1nd has the capacity to make smaller, less intelligent models perform like frontier models. By feeding the LLM only the exact, causal, and structurally relevant nodes instead of dumping files into context, small open-source models succeed where large proprietary models drown in noise.
Neuro-Symbolic Editing: grep just searches text. m1nd applies edits atomically and instantly runs a topological graph verification (verify=true). It re-ingests the changed files to mathematically prove whether the agent's code broke dependencies, introduced syntax errors, or worsened the graph's coherence before the AI takes another step.
Sub-Millisecond Speed: Pure Rust. Codebases with tens of thousands of nodes compute activations in microseconds. In one real case, a memory pool of 259,000 nodes was queryable via RAM in under 1ms.

335 files -> 9,767 nodes -> 26,557 edges in 0.91 seconds.
Then: activate in 31ms. impact in 5ms. trace in 3.5ms. learn in <1ms.

Proven Results

Live audit on a production Python/FastAPI codebase (52K lines, 380 files):

| Metric | Result | |--------|--------| | Bugs found in one session | 39 (28 confirmed fixed + 9 high-confidence) | | Invisible to grep | 8 of 28 (28.5%) -- required structural analysis | | Hypothesis accuracy | 89% over 10 live claims | | Post-write verify accuracy | 100% — 12/12 scenarios (SAFE/RISKY/BROKEN) | | LLM tokens consumed | 0 -- pure Rust, local binary | | m1nd queries vs grep ops | 46 vs ~210 | | Total query latency | ~3.1 seconds vs ~35 minutes estimated |

Criterion micro-benchmarks (real hardware):

| Operation | Time | |-----------|------| | activate 1K nodes | 1.36 µs | | impact depth=3 | 543 ns | | flow_simulate 4 particles | 552 µs | | antibody_scan 50 patterns | 2.68 ms | | layer_detect 500 nodes | 862 µs | | resonate 5 harmonics | 8.17 µs |

Quick Start

git clone https://github.com/maxkle1nz/m1nd.git
cd m1nd && cargo build --release
./target/release/m1nd-mcp

// 1. Ingest your codebase (910ms for 335 files)
{"method":"tools/call","params":{"name":"m1nd.ingest","arguments":{"path":"/your/project","agent_id":"dev"}}}
// -> 9,767 nodes, 26,557 edges, PageRank computed

// 2. Ask: "What's related to authentication?"
{"method":"tools/call","params":{"name":"m1nd.activate","arguments":{"query":"authentication","agent_id":"dev"}}}
// -> auth fires -> propagates to session, middleware, JWT, user model
//    ghost edges reveal undocumented connections

// 3. Tell the graph what was useful
{"method":"tools/call","params":{"name":"m1nd.learn","arguments":{"feedback":"correct","node_ids":["file::auth.py","file::middleware.py"],"agent_id":"dev"}}}
// -> 740 edges strengthened via Hebbian LTP. Next query is smarter.

Add to Claude Code (~/.claude.json):

{
  "mcpServers": {
    "m1nd": {
      "command": "/path/to/m1nd-mcp",
      "env": {
        "M1ND_GRAPH_SOURCE": "/tmp/m1nd-graph.json",
        "M1ND_PLASTICITY_STATE": "/tmp/m1nd-plasticity.json"
      }
    }
  }
}

Works with any MCP client: Antigravity, Claude Code, Codex, Cursor, Windsurf, Zed, or your own.

For large codebases, see Deployment & Production Setup for how to run m1nd as a persistent server with smart namespace ingest and near-zero latency.

The Context Window Is Not Intelligence

The dominant approach in AI-assisted development treats the context window as a substitute for understanding: inject raw files, search by text pattern, let the LLM infer structure. This works at small scale and fails at the scale of real codebases.

m1nd is built on a different premise: code structure is a graph problem, not a text problem.

A neuro-symbolic connectome — built in pure Rust, running locally, persisted across sessions — encodes every function, module, type, and file as a node with weighted edges. Spreading activation, Hebbian plasticity, and structural hole detection replace grep, glob, and linear file reads.

The practical consequences are measurable: 84% fewer tokens consumed per session, sub-millisecond graph traversal, and the detection of bugs that exist in the absence of code — structural holes that no text search engine can find because there is no text to match.

We designed m1nd after observing that the context window paradigm, while commercially convenient, architecturally penalizes agents for understanding codebases. Every file loaded is a cost. Every grep is an approximation. m1nd removes both constraints.

If you believe code intelligence should be mathematically rigorous — graph topology over pattern matching, structural truth over probabilistic guessing — build on this. Star the repo, open issues, contribute. The architecture is open.

It worked? Star this repo -- it helps others find it. Bug or idea? Open an issue. Want to go deeper? See EXAMPLES.md for real-world pipelines.

Configure Your Agent

m1nd is designed to replace grep, glob, and blind file reads fo

M1nd

Install / Use

README

Architecture & Paradigm Shift

Proven Results

Quick Start

The Context Window Is Not Intelligence

Configure Your Agent