Introduction

Ogcode is an agent operating environment for software work. Most AI coding tools answer a question or complete a line; Ogcode is built to take responsibility for multi-step work — understanding a codebase, making a bounded change across files, running the project’s own tests, and leaving behind evidence you can review.

If you have used an editor plugin with a chat panel, the difference is the shape of the loop. Ogcode gives an agent a workspace, a set of tools, a budget, and a stopping condition, then lets it run until the work is done or it needs you.

What it is

• Open source and self-hostable. One binary with the web UI built in — nothing else to serve, no database to provision.
• Agentic, not autocomplete. The agent plans, edits, executes and verifies, rather than proposing a diff you paste in yourself.
• Browser-native. You work in a web UI, not inside a particular editor.
• Model-agnostic. Anthropic, OpenAI, OpenRouter, a local Ollama endpoint, or the OG Lab subscription (OGX).
• Built for long-running work. Sessions survive restarts, context is compacted rather than lost, and memory carries from one turn to the next.
• Git-native. Plan Mode breaks a feature into tasks, runs each in its own worktree, and opens pull requests.
• Permission-aware. Every consequential action is gated — ask, auto-approve, or refuse by an explicit rule.
• Growing into a general computer agent. Code is the current focus; the same tool-and-permission model reaches documents, rich output, and local services.

What you can do with it

• Understand a codebase. Ask how a feature works and get an answer grounded in the real files, pointed straight at the parts that matter.
• Build and fix. Describe a change and the agent edits the files, runs the build and tests, and reports what it changed and how it verified it.
• Plan larger work. Give it a whole feature and let Plan Mode split it into tasks it can run in parallel and turn into pull requests.
• Research. Let it search the web and read pages alongside your local code and documents.
• Repeat your process. Teach it reusable skills, and connect the tools you already use.
• Work headlessly. Run any of it from a script or CI with a single command.

Where to go next

1. Install — get the binary onto macOS, Linux or Windows, or run it in Docker.
2. Quick start — go from a fresh install to an agent editing files in your repo.
3. Core concepts — sessions, modes, memory and permissions, the mental model behind everything else.

The rest of the site is a set of guides, one per capability, when you want to go deeper.