prasenjeet-symon/ogcode: aI agent that understands your codebase, uses real tools, researches the web

The open-source, self-hostable AI coding agent.

Ogcode is an AI agent that understands your codebase, uses real tools, researches the web, remembers decisions, plans complex work, and ships changes — from your browser. Your computer, your models, your data, your rules.

Installation

curl -fsSL https://ogcode.in/install.sh | sh

# Package managers
brew tap prasenjeet-symon/tap && brew install ogcode   # macOS and Linux
winget install prasenjeet-symon.ogcode                 # Windows
go install github.com/prasenjeet-symon/ogcode@latest   # any OS with Go

[!TIP] CGO_ENABLED=1 is required when building from source — the Swift tree-sitter binding is cgo.

On Windows without winget:

irm https://ogcode.in/install.ps1 | iex

Or run the container:

docker run -p 9595:9595 \
  -v ~/.ogcode:/root/.ogcode \
  -v "$(pwd):/workspace" -w /workspace \
  ghcr.io/prasenjeet-symon/ogcode:latest

The image is also on Docker Hub as prasenjeetsimon/ogcode:latest.

Start it and open the web interface:

ogcode            # serves the UI on http://localhost:9595

Use a local model with Ollama — no API key needed:

ollama serve
ogcode

Permissions are per session: Ask (approve each step), Auto (risk-gated), or Yolo (no prompts). Learn more about permissions.

What you can do with Ogcode

• Understand unfamiliar code — Ogcode maps a project before reading it, outlines files with exact line ranges, and reads only what matters. Core concepts →
• Build and fix software — implement features and fixes across files, run your builds and tests, and investigate failures instead of retrying blindly.
• Deliver a whole feature — turn an objective into tasks, run independent ones in parallel, and open pull requests for review. Plan mode & tasks →
• Stay in control — choose Ask, Auto, or Yolo per session, and approve or remember each decision. Permissions →
• Keep context that fits — recall past decisions instead of replaying the whole transcript, so long sessions stay focused and affordable. Memory & context →
• Research as you go — built-in web search and page reading, reusable SKILL.md workflows, and external MCP servers. Search, skills & MCP →
• See more than text — Mermaid diagrams, LaTeX and rendered PDFs, Plotly charts, sandboxed HTML, and downloadable artifacts. Rich results & preview →

Documentation

For configuration, providers, skills, remote deployment, and everything else, head over to our docs.

Configuration

Ogcode detects your provider from the environment. Set at least one:

Settings can also live in ogcode.json at the project root (project settings) and ~/.config/ogcode/config.json (global settings); environment variables override both.

{
  "providers": {
    "ollama": { "baseUrl": "http://localhost:11434" },
    "anthropic": { "apiKey": "sk-ant-..." }
  },
  "skills": {
    "paths": ["./team-skills"],
    "permissions": { "deploy-prod": "ask" }
  }
}

OGX is a subscription plan from OG Lab. Connect it from the settings screen and the plan's models run through OG Lab's gateway — no environment variable needed.

Remote deployment and security

Ogcode can run on a remote machine and be reached from a browser — but it can read and modify files and run commands. Never expose it directly to the public internet without authentication.

Recommended boundaries:

1. Bind to localhost and reach it over an SSH tunnel.
2. Put a reverse proxy with HTTPS and authentication in front of it.
3. Use a VPN such as WireGuard or Tailscale.
4. Run high-risk work in Docker, a VM, or an isolated worker.

ssh -L 9595:localhost:9595 user@your-server

Working setups — SSH tunnel, reverse proxy, and Docker — are in the remote deployment guide. For a hosted, multi-user deployment, see the control-plane documentation.

Security

Treat Ogcode like a powerful automation process:

• Review permissions before enabling write or shell access.
• Keep API keys and secrets out of prompts, repositories, and public artifacts.
• Use isolated environments for untrusted code.
• Do not expose an unauthenticated server to the internet.
• Report vulnerabilities privately through the project's security channels.

Your code and session data stay local; only conversation content is sent to the model provider you configure, where that provider's privacy policy applies.