Overview

An agent is the unit you talk to in Nairi. Each agent runs on Nairi infrastructure and answers prompts via Claude, Codex, OpenCode, or Cursor. Agents are identified by a human-readable slug (the agent_id, for example my-agent) — that slug is what you pass throughout the rest of the API.

How it works

Agents have two layers:

• Configuration — name, description, system prompt, instance count, and attached resources (rules, MCP configs, skills, vaults, env vars). All managed via this API.
• Deployment — the actual running agent. Created with POST /agents/{agent_id}/deploy, which is asynchronous and returns immediately while the work happens in the background.

When to use it

• Stand up a shared agent for your team in Slack or Discord. Define its prompt and tools once, attach it to a channel, and let teammates @-mention it from where they already work.
• Provision a dedicated agent per customer or tenant. Spin one up at signup, tear it down at churn — no manual dashboard clicks.
• Roll config changes without restarting containers. Update the system prompt or swap attached resources, then call /deploy with update_config_only: true.

Related

• Agent resources — attach rules, MCP configs, skills, vaults, and integrations to an agent.
• Env vars — plaintext variables injected into the agent's container.
• Conversations — once an agent is deployed, you drive its work through conversations.