Ogden — Setup Guide

1

Prerequisites

Docker and Docker Compose, installed and running — that's genuinely the only thing your machine needs. There's no separate Node, Python, or database install; everything runs inside containers.

You'll also need your license: accept the private GitHub repository invite you received after checkout (check your email, including spam — GitHub sends this, not us), then clone it:

# from wherever you keep your projects
git clone https://github.com/<your-invited-repo>.git
cd <your-invited-repo>

2

Configure your environment, then start it up

Copy the example environment file. You can leave it entirely blank for now — every setting it covers (AI provider keys, admin credentials, encryption keys) can also be set later from inside the app or is generated automatically on first boot.

cp .env.example .env
docker compose up --build

First boot pulls base images and builds the app, so give it a minute or two. Once it settles, the frontend is at http://localhost:8501 and the backend API at http://localhost:8000.

3

Sign in

Ogden creates one organization and one owner login the first time it boots. If you set OGDEN_ADMIN_EMAIL/OGDEN_ADMIN_PASSWORD in .env before that first boot, those are your credentials. Otherwise, a random password is generated and printed once to the backend's logs:

docker compose logs backend

Write it down — it isn't shown again. You can change it from the sidebar any time.

4

Connect an AI provider

Ogden doesn't come with an AI included, and it doesn't ship any API keys or credits of its own — it's the crew's workspace, not the crew's brain. You bring the model. On the Providers page, add a connection to whichever you'd like: OpenAI, Anthropic, Google Gemini, a self-hosted Ollama instance, or any OpenAI-compatible endpoint. Different personas in the same project can each use a different one.

Prefer to keep everything fully local? Point a connection at an Ollama instance and nothing about your prompts, code, or output ever leaves your own infrastructure. Every other option sends data to that provider directly from your Ogden instance under that provider's own terms — see the Privacy Policy for more on this.

5

Build your team

On Team Builder, create a project — start from the default 7-agent template (a Product Manager, a couple of engineers, QA, DevOps, and an Executor that actually runs tools) or build your roster from scratch. Every persona is yours to tune: display name, avatar, system prompt, which connection and model it uses, temperature, budget cap, and which sandboxed tool groups it can call.

6

Load the backlog

On the Board, add a task and assign it to a persona. Flip the project to Autonomous and it gets picked up and worked on its own, or leave it manual and click Run now whenever you want to drive it yourself. Turn on Require approval and every run lands on its own branch for you to review before it merges.

7

Watch it work

Team Chat talks to the whole crew at once; Agents pulls one persona aside 1:1 and shows their task/run history. Code shows real commit history, diffs, and a branch review view — no need to check anything out locally just to see what changed. Memory is the crew's long-term recall: search, add notes by hand, or let it capture task outcomes automatically.

8

Running the Business tier

Same codebase, same image — DEPLOYMENT_TIER just turns on real multi-user accounts, role-based access, and Postgres instead of SQLite for concurrent writes. Set POSTGRES_PASSWORD in .env first, then layer the Business compose file on top of the base one:

docker compose -f docker-compose.yml -f docker-compose.business.yml up --build

Once you're signed in, the Team page lets an Admin or Owner add members and set their role — Viewer (read-only), Member (works within projects), Admin (manages projects, connections, and other members), or Owner (all of that, plus managing other Owners).

9

The rest of Configure

A few more pages worth knowing about, all optional: MCP Servers lets a persona discover and call tools from any Model Context Protocol server you register. Notifications sends an alert (webhook, Slack, Discord, or email) when a budget's hit, a task's blocked, or a run fails. Settings covers branding (handy if you're running more than one instance), the autonomy concurrency limit, and — once you've connected a provider that supports it — semantic memory search and cache routing.

10

Getting help

Your license includes a year of updates and support from the date of purchase — see the Terms & Conditions for the full renewal terms. Questions about setup, deployment, or your license: hello@harrisonsmith.ai.

Deploying somewhere other than your own machine? Set CORS_ALLOWED_ORIGINS in .env to the real origin your frontend will be served from — see the comment above it in .env.example for the exact format.

Get your crew running.