Graphicle: Rust Port of Graphify Adds Memory and MCP Integration for Coding Agents

Overview

Graphicle is a Rust-based port of Graphify designed to enhance coding agents by combining code graph mapping with persistent project memory. It operates locally without requiring API keys, accounts, or telemetry, supporting macOS and Linux (with Windows support planned). The tool is compatible with major AI coding environments including Claude Code, Cursor, Codex, VS Code, and Claude Desktop via the Model Context Protocol (MCP).

Key Features

• Unified Architecture: Unlike previous setups requiring separate code-graph tools and memory servers, Graphicle uses a single installation, a single SQLite file for storage, and one MCP server. This eliminates the need for Python, Docker, or external database servers.
• Code Graph & Memory: It maps repositories into code graphs using tree-sitter parsing (no model calls required) and stores project documents, decisions, and past agent sessions as searchable memory.
• Cited Context: The graphicle context command and MCP tools return excerpts with file and line-number citations within a token budget, ensuring agents have grounded evidence.
• Session Capture: Completed agent turns are saved, and approved memories are loaded when new sessions start. Memories are anchored to specific functions and marked stale if the underlying code changes.

Performance Benchmarks

The article reports performance comparisons against the original Graphify tool:

• Kubernetes Repository (~31,000 files): Graphicle 0.1.0 engine completed extraction in 3 minutes 18 seconds, compared to 26 minutes 38 seconds for Graphify 0.9.88. Both tools built identical graphs (256,329 nodes, 1,400,392 edges).
• Django Repository: Graphicle pre-release build took 7.5 seconds for code-only extraction, versus 72.8 seconds for Graphify 0.9.77 (median of 5 alternating runs).
• Memory System: In benchmarks against 7 other memory systems using a single open model as judge, Graphicle's pre-release build did not trail any competitor across three development sets.

Privacy and Setup

• Local-First: Data is stored locally in a single SQLite store. Network usage is limited to loopback by default; external API calls only occur if users explicitly configure models (Anthropic, OpenAI, Ollama) or use specific features like full document extraction.
• Security: The tool respects .gitignore, skips credentials, and masks sensitive data in session captures. Deletion of memory or sessions also removes derived data, with hash-chain receipts provided.
• Installation: Available via brew install graphicle-ai/tap/graphicle or an install script. The graphicle setup command automates indexing, agent client connection, and Git hooks installation, with graphicle disconnect available for reversal.

Licensing

Graphicle is released under the Apache-2.0 license, allowing free personal and commercial use. It credits Safi Shamsi and the Graphify contributors for the underlying code graph engine.