Create and interact with a Trinity agent using the Web UI, API, or MCP tools.
📺 Watch: Build an AI Recruiter Agent — zero to deployed (Jun 2026) · From Zero to Deployed AI Agent (Apr 2026) · all videos
The fastest path: let Trinity guide you. On a fresh install with no agents yet, a short onboarding wizard opens automatically the first time you reach the Dashboard after logging in. It asks one question — what do you want your first agent to do? — then:
- Opens the Create Agent form with a matching starter template pre-selected.
- After the agent is created, walks you to the Claude subscription step so your agent can actually think (Settings → Integrations → Claude Subscriptions), or straight to chat if Claude auth is already configured.
You can dismiss it at any time ("Skip for now") — it won't nag you again.
Relaunch the wizard any time (e.g. to spin up another agent, or if you skipped it):
open the Dashboard with the ?onboarding=1 query parameter:
http://localhost/?onboarding=1
This works regardless of how many agents you already have. Log in first, then
open the link (opening it while signed out sends you through the login page, which
drops the ?onboarding=1 parameter). On a fresh, empty install the Dashboard also
shows a Get started button in the empty state that opens the same wizard.
Prefer to do it manually? Skip the wizard and follow How It Works below.
- Open http://localhost and log in as admin.
- Click Create Agent on the Dashboard or Agents page.
- Choose a template:
- GitHub template -- Select from configured repos (e.g., pre-built agent templates).
- From scratch -- Creates a minimal agent with a default CLAUDE.md.
- GitHub URL -- Paste
github:Org/repoorgithub:Org/repo@branchfor a specific branch.
- Enter an agent name (lowercase, hyphens allowed).
- Click Create -- Trinity clones the template, builds the container, and starts it.
- Navigate to the agent's detail page to interact via Terminal, Chat, or Tasks.
- A Docker container is built from the
trinity-agent-baseimage. - Template files are copied into the agent's workspace at
/home/developer/. - Credentials from
.mcp.json.templateare detected and shown as "missing" until configured. - The agent starts automatically and appears on the Dashboard.
- Terminal tab -- Direct Claude Code TUI access (Claude/Gemini/Bash modes).
- Chat tab -- Simple chat bubble interface.
- Tasks tab -- Send one-off tasks and view execution history.
- Files tab -- Browse and edit agent workspace files.
- Add credentials in the agent's Credentials tab.
- Send a task via the Tasks tab or chat via the Chat tab.
- Set up schedules for autonomous operation.
- Configure permissions if building multi-agent systems.
# Get auth token
TOKEN=$(curl -s -X POST http://localhost:8000/api/token \
-d 'username=admin&password=YOUR_PASSWORD' | python3 -c "import sys,json; print(json.load(sys.stdin)['access_token'])")
# Create agent from template
curl -X POST http://localhost:8000/api/agents \
-H "Authorization: Bearer $TOKEN" \
-H "Content-Type: application/json" \
-d '{"name": "my-agent", "template": "github:Org/repo"}'create_agent(name="my-agent", template="github:Org/repo")
- Setup -- Initial platform installation and configuration.