Graphide · closed alpha
Download Graphide
The code editor that maps what your agents are building.
Downloads need a Graphide account: Sign up / Sign in.
Or map a public GitHub repo in your browser, nothing to download.
The problem
Agents write code faster than anyone can read it.
One prompt can touch forty files. You get two thousand lines of diff and a choice: read every line, or trust it. Either way, the design in your head and the code on disk drift apart, and nothing shows you where.
DiffForty-one files, about two thousand lines. The change to auth/session.go is in there, past line one thousand.
Shown as a diff: 41 files and 1,996 changed lines.
How it works
One graph. You, your agents and the code all work on it.
gr init reads every file and builds the graph: modules, files, classes and functions, the calls between them, and a plain-English description of each node. Shown here on Flask: all 1,882 of its symbols, one dot each, from Graphide’s own extraction.
Built for real codebases
Fits the codebase you already have.
No rewrite, no new language, no migration. We have run Graphide on fzf, Flask, Hono and Excalidraw; point it at the repository you have today.
Reads the languages you ship
Go, TypeScript, JavaScript, Python, Java, Elixir, SQL, Nix, HTML, CSS, Circom. Every symbol becomes a node with a one-line description of what it does.
Works with your agents
Claude Code and Codex run in the editor’s Agents panel. The graph is also an MCP server, so any agent that speaks MCP can search it and edit it.
Lives in your repo
The graph is versioned with your code. It commits, diffs and merges alongside it, so a pull request carries its architecture.
Undo anything
Every change to the graph has history and can be undone, one edit at a time or a whole build at once.
Zoom from system to symbol
Start at the modules and the dependencies between them. Open one to see its files, then its functions, without losing your place.
Scriptable from a shell
gr init builds the graph, gr sync keeps it current and gr check verifies it, so a git hook or a CI job can hold the line.
Why we are building it
In the future, people won’t read codebases. They’ll read maps of them.
Agents already write code faster than anyone can review it, and they keep getting faster. We think that before long nobody will work through a codebase line by line. They will read a picture of it, what the parts are, how they connect and what just changed, and steer from there. Graphide is that picture, and you can edit it.
It started with Mermaid diagrams. Keeping AI-written code from turning into slop meant spending most of the day drawing the architecture by hand and checking agent output against it. The code worked; it just kept drifting from the design, and nothing showed where.
Dan McGee
CEO, co-founder
Runs the cloud, agent routing and model orchestration.
Benjamin McIntyre
CTO, co-founder
Runs the graph, the compliance checks and the local daemon.
Dan and Ben met in middle school in Cincinnati and have been building things together ever since.
Ben has been writing editor tools since then, and at the University of Cincinnati started the Linux and open-source club, which grew past a hundred members.
Dan ran a small automation software business for a few years in high school and got hooked on startups. That led to Berkeley’s M.E.T. program (EECS and business), then to Illumio, building Matrix, which simulates attacks across networks of 300+ virtual machines.
Now they both write most of their code with agents, and Graphide is the editor they wanted for it. Graphide is backed by Y Combinator (F26) and based in San Francisco.
Questions
Before you point it at your code.
What is Graphide?
A code editor that helps you keep up with what your AI agents are building. It turns your codebase into a live map of its architecture, every module, file, class and function and every call between them, that you and your agents can read and edit. It is a full editor, built on VS Code.
Who is it for?
Developers shipping production code with AI agents, especially on codebases too big to hold in your head. If you have ever scrolled through a two-thousand-line agent diff hoping nothing important was in it, it is for you.
Which agents does it work with?
Claude Code and Codex run inside the editor's Agents panel. The graph is also served over MCP, so any agent that speaks MCP can search it and edit it.
Which languages does it read?
Go, TypeScript, JavaScript, Python, Java, Elixir, SQL, Nix, HTML, CSS and Circom.
Where does my code go?
The graph is built on your machine, from source files Graphide reads there. While you are signed in, Graphide continuously syncs the graph itself (symbol names, function signatures, short excerpts of your code and file paths) to its cloud, which is what runs the AI features (node descriptions, the architecture view and compliance checks). Telemetry is on by default at the full level, which also sends us snippets of your sessions, each cut to at most 4 KB: your prompts, your agents' replies and tool calls with their results, and your edits. Those snippets can include your code. The diagnostic level leaves them out and sends only crash and usage data; off sends only a usage ping. When an agent or an analysis needs code, it goes to a model provider. The privacy policy lists exactly what is sent, to whom, and for how long. Read the privacy policy.
What does it cost?
Nothing during the alpha.
Which platforms?
macOS, Windows and Linux. macOS builds are signed and notarized. Windows builds are not signed yet, so Windows may show a SmartScreen warning on first launch.
How do I get in?
Request access and create an account with GitHub or Microsoft. We let people in in small batches while the alpha is small. If someone gave you an access code, enter it when you sign up and you are in straight away.
Something else? Write to hello@graphide.com. Two founders read every one.
Closed alpha
Bring your own codebase.
Graphide is invite-only while we work closely with our first users. Request access and we will let you in; an access code skips the queue.