Who it is forA privacy-minded developer or tinkerer who wants to try an agent workspace without sending prompts or files to a cloud model.
The problemAgent platforms usually assume a paid cloud API, which costs money and sends your data off the machine.
The outcomePilot Deck using a model served by Ollama at localhost:11434, configured in one YAML file, for small local tasks.
What you need first
- Node.js v22.13.0 or newer and below v23
- pnpm via corepack
- A model provider key, or Ollama for local models
Steps
- Install Pilot Deck and Ollama
- Start Ollama with
ollama serve - Pull a model:
ollama pull qwen3:0.6b, or a larger one your hardware can run - In ~/.pilotdeck/pilotdeck.yaml set agent model to ollama/qwen3:0.6b and add an ollama provider with protocol openai and url http://localhost:11434/v1
- Restart Pilot Deck with
pilotdeckand open http://localhost:3001 - Run a small task in a new WorkSpace and check the replies come from the local model
Prompt for your agent
Configure OpenBMB/PilotDeck to use Ollama with <model>, following the README's ollama YAML example. Restart it, run a short test task, and tell me whether the model is fast enough on this machine.
Caveat. qwen3:0.6b is tiny; long or complex tasks need a much larger model.
Use this in your AI agent
Ask your agent
Using the Deduped MCP, get the setup steps for OpenBMB/PilotDeck and "Run the agent platform fully on a local model"
Add Deduped to Claude Code once
claude mcp add --transport http deduped https://dedupedai.com/mcp
Every Monday: the 5 hottest repos and exactly how to use them. Free, one email a week, unsubscribe anytime.
Get them freeShare: X LinkedIn Hacker News Email Copy link
Third-party repos we don't own. AI-written use cases can be wrong. Disclaimer