How Do You Set Up OpenBot AI Coworkers? Clone and Run

Intro — what this is

OpenBot is an open-source, alpha template from CopilotKit (github.com/CopilotKit/OpenBot), not an npm package: you clone the repo, copy .env.example to .env, add a CopilotKit Intelligence key and an OPENAI_API_KEY, then run bun install and bash scripts/start.sh with Docker running. The script starts Postgres and the Bot computers, the API on port 3001 and the app on port 3010, and every Bot action then passes through one CEL policy gateway that writes an audit row before it acts.

Upstream describes the project as "a template, not a product": no hosted signup product in that framing, nothing published as a package you npm install and forget. You clone the repo, replace the example coworkers under examples/ with your own, and run it yourself. The GitHub repo marks it alpha and under active development.

Related links:

• 01shoppercove.comDeploying DeepSeek Harness routing suite
• 02shoppercove.comAbout ShopperCove
• 03github.comCopilotKit/OpenBot on GitHub
• 04copilotkit.aicopilotkit.ai/openbot

Honesty bar: This post is a first-look / orientation rewrite so the live CMS page stops showing a raw README dump. A deeper ShopperCove policy-gateway lab (CEL deny-path, engine evidence, Docker limits) lives in a separate package and should get its own URL when published — do not confuse the two. Affiliates: 0.

Install (clone path)

There is no package to npm install. You clone the repository and bring it up with Docker Compose and Bun. Documented quick-start shape:

# 1. Create your env file
cp .env.example .env

# 2. Get CopilotKit Intelligence credentials
npx --yes copilotkit@latest login
npx --yes copilotkit@latest project select
# put the cpk-... key from `project select` into .env as INTELLIGENCE_API_KEY

# 3. Fill the remaining required model key in .env:
# OPENAI_API_KEY

# 4. Install and run
bun install
bash scripts/start.sh

Requirements per upstream: Docker (PostgreSQL + Bot computers), Bun, a CopilotKit Intelligence project (a free plan exists; Intelligence can be self-hosted), and a model key — the proof-of-concept Bot uses OpenAI; the LangGraph Bot can use OpenAI, Anthropic, or Google.

scripts/start.sh starts Docker services, applies migrations, starts the API on port 3001, starts the app on port 3010, and checks health routes before printing next steps. Then open Open localhost.

The repo also ships a prompt.txt meant to be pasted into an AI assistant for setup help — same steps, with notes on which of the many blank keys in .env.example you actually need to fill yourself versus which the start script generates.

Hello world

Once the app is running, the shortest documented interaction is: open /bot and give it a task.

Open news.ycombinator.com and tell me the top story.

No agent configuration is required first — a default Bot is already wired up. Useful follow-ups from the docs:

• Ask the Bot to fill out Open httpbin.org, then inspect /admin/audit for the recorded action trail.
• Open /admin/boundaries, add a deny rule, and retry the same browser action to watch it get refused.

The one concept you must understand first

Everything in OpenBot routes through one gateway, and that gateway is the point of the project.

Upstream framing: anything a Bot does to a computer, a file, an MCP server, or a component goes through one gateway that decides and records it. That is the difference between an agent that can use your tools and an agent you can let near them.

Concretely, the documented sequence is:

1. Resolve the target from a server-held snapshot (not model prose about what it clicked).
2. Evaluate a CEL policy (deny checked before allow; a missing policy permits nothing; a broken rule refuses rather than opens).
3. Write an audit row.
4. Only then call the Bot's computer — no path that acts before the record exists.

Once you see the gateway as a mandatory checkpoint rather than a logging layer bolted on afterward, boundaries, audit, credentials, and MCP grants read as configuration of that one checkpoint.

What this first look does not claim

If you need the measured deny-path notes (pinned commit, unit tests, harness allow→deny with rule citation, Docker hard-stop), wait for / publish the dedicated gateway hands-on post — do not paste that lab into this slug.

Next steps (upstream docs)

• Full docs index past quick start: docs/README.md in the repo
• Deployment sizing / multi-replica notes: docs/deployment.md
• AG-UI protocol any Bot must speak: ag-ui-protocol/ag-ui
• Product landing: copilotkit.ai/openbot

Related ShopperCove AI-tooling (different layer — Harness routing presets, not coworker computers):

• 01shoppercove.comDeploying DeepSeek Harness routing suite

FAQ

Is OpenBot a hosted SaaS I can sign up for?

Upstream frames a self-hosted template you clone. Do not expect a "sign up and get coworkers" product from the README framing we cite.

Why Bun and Docker?

Upstream's documented path uses Bun for the app/API and Docker for Postgres + Bot computers. Missing Docker is a hard stop for scripts/start.sh.

What is CEL doing here?

Common Expression Language rules on fields like page.host, tool.name, file.*, mcp.*. Deny beats allow; empty/missing policy fails closed.

How does this relate to DeepSeek Harness posts?

Different product class: OpenBot = governed coworker computer; DeepSeek Harness = agent IDE / routing presets. Link for cluster discovery; do not merge.

Should I treat alpha as production?

No. Verify boundaries and audit on your clone before handing a Bot real credentials.

Verdict

OpenBot's interesting idea is not "another chat UI" — it is resolve → CEL → audit → act in front of a real browser and filesystem. This first-look post is enough to decide whether to clone. It is not enough to claim ShopperCove finished a live deny-path UI lab on this URL. Keep scopes separate when the deeper gateway post goes live.

Affiliates: 0. No ClickBank. API/model spend is a cost disclosure, not an affiliate pitch.

FAQ

Where is OpenBot on GitHub?

At github.com/CopilotKit/OpenBot. The repo marks it alpha and under active development, and upstream calls it a template, not a product.

What do I need to run OpenBot?

Docker (for PostgreSQL and the Bot computers), Bun, a CopilotKit Intelligence project (a free plan exists) and a model key. The proof-of-concept Bot uses OpenAI.

What does the OpenBot gateway do?

It resolves the target from a server-held snapshot, evaluates a CEL policy (deny beats allow; a missing policy permits nothing), writes an audit row, and only then calls the Bot's computer.

Related ShopperCove posts

• 01shoppercove.comDeploying DeepSeek Harness routing suite
• 02shoppercove.comLily in CI: Trying Fuzz-Based Backdoor Detection on a Real Repo
• 03shoppercove.comCosign + SBOM in CI
• 04shoppercove.comAbout ShopperCove

External sources (links only — no README paste)

• 01github.comCopilotKit/OpenBot
• 02github.comOpenBot architecture.md
• 03github.comAG-UI protocol
• 04copilotkit.aicopilotkit.ai/openbot