Skip to content
Claude Control Documentation home
InboxAsk a human

Deploying your first agent

An agent in Nairi is a shared AI teammate. You deploy it once and everyone in your workspace talks to the same agent, so there are no per-seat licences and no terminal setup for each person.

Before you start

You need access to app.nairi.ai and an account with one LLM provider. Everything else is optional.

1. Connect an LLM provider

Go to Settings → Integrations and connect one of:

  • Anthropic, which gives you the Claude Code harness and Claude models.
  • Codex, which gives you OpenAI's Codex harness and models.
  • OpenCode Zen, which gives you the OpenCode harness and models from more than 30 providers.

This is the only required integration. Slack, Discord and GitHub can wait.

2. Connect a chat platform (optional)

Connect Slack or Discord if you want your team to talk to the agent in a channel. Skip it if you only need the web app or the API.

Connect GitHub if the agent should read your code, push branches and open pull requests, or if you keep your skills, MCP configs and AGENTS.md in a configuration repository.

3. Create the agent

Open Fleet and choose New agent. Two fields are required:

  • Name, which is what people @mention and what the API uses. Keep it short and unique.
  • LLM integration, chosen from what you connected in step 1.

You can also attach a repository, rules, skills, MCP configs and environment variables now, or add them later. The system prompt comes pre-filled with Nairi's defaults, so leave it alone unless you have a reason to change it.

Choose Managed hosting to have Nairi run the agent, or Self-hosted to run it on your own machine at no cost. Then choose Deploy and wait for the agent to come online.

4. Send it a prompt

In Slack or Discord, mention the agent at the start of your message: @your-agent summarise yesterday's incidents. In the web app, use the Tasks dashboard. Every transcript and every tool call is recorded under Tasks, so you can always see what the agent actually did.

What next

Set up a scheduled job if you want a prompt to run on its own every morning. Share the agent with teammates. The playbooks show four worked examples, including a code reviewer and an AI SRE.