Graphicle brings fast code graphs and memory to coding agents

A developer has released Graphicle, a program that maps code repositories into graphs and stores project memory, then feeds that information to AI coding agents. It is a portable rewrite of the older Graphify tool, built in the Rust programming language for speed.

The tool works with Claude, Codex, Cursor, Microsoft VS Code, Claude Desktop and any client that supports the Model Context Protocol, known as MCP ¹. The idea is simple: instead of guessing what your codebase looks like, your agent gets exact file paths, line numbers and cited excerpts when you ask it to find something.

Graphicle builds a code graph by parsing files with tree-sitter, an open-source parsing library, without using any AI model ¹. It also records project documentation, reviewed decisions, procedures and past agent sessions. All of that lives in one SQLite database on your machine. There is no account, no API key and no data sent back to a server. The only automatic network request is a daily check for new software releases, which can be turned off .

Speed advantage

The main pitch is performance. In early tests, the 0.1.0 version processed a repository with 31,000 files in 3 minutes and 18 seconds, compared with 26 minutes and 38 seconds for Graphify 0.9.88 on the same machine with the same number of processors ¹. The tool claims to tie grep, a classic command-line search program, on locating code .

When an agent asks a question, Graphicle returns relevant excerpts from code and documents within a token budget set by the user ¹. Every excerpt includes its file name and line range so the result can be checked directly. Answers require a configured model like Anthropic's API, but the evidence gathering itself does not call any AI model .

Memory tied to code

Graphicle tracks memories about a project and links them to the functions they describe. If a function changes, the associated memory is marked as potentially stale, while edits elsewhere in the same file usually leave it current ¹. Reviews of memories must be approved by a human before the agent can use them .

Sessions are captured turn by turn. Once memories are reviewed, each new agent session begins with those memories plus the latest conversation turns ¹. The tool can also work across repositories in a workspace, letting agents draw on code from multiple projects at once .

Graphicle installs by running a setup command in a repository. It indexes the code, builds the graph, connects any supported coding agents it finds, adds Git hooks to keep things current and appends two lines to an AGENTS.md file unless Claude Code is the only agent in use ¹. It backs up every file before changing it, and a disconnect command reverses the entire process . A web app lets users browse the graph, read sources and approve memories .

The project is open source on GitHub and supports installation through Homebrew, a direct install script and a preview Windows build ¹.

Key facts

• Graphicle is a Rust port of Graphify that builds code graphs and stores project memory locally
• On a 31,000-file repository it ran a code extract in 3 minutes 18 seconds versus 26 minutes 38 seconds for Graphify
• It works with Claude, Codex, Cursor, VS Code and other MCP-compatible clients
• No API key, account or telemetry required; all data stays on the user's machine
• Memories are linked to code functions and expire when that code changes

Sources

1. GitHub - Graphicle-AI/graphicle: Code graph and memory for your coding agents, in one Rust program. For Claude Code, Cod