Tutorial: set up Agentic and complete your first task¶
This tutorial takes you from an empty machine to a task with an agent-written summary and a completed agent run. It assumes no prior knowledge of Agentic.
What you will build¶
A running Agentic app on your own machine, one task that points at a registered repository, a git worktree for that task, and an agent run whose transcript streams into the app.
Prerequisites¶
Install these first:
- Python 3.12 or newer.
- uv.
- Node.js 20 or newer.
- At least one ACP harness on your
PATH, eitheropencode/opencode2orpi-acp.
Clone the repository and move into it:
Step 1: install the backend and CLI¶
uv sync creates a virtual environment and installs the backend (agentic-server) and the CLI (agentic).
Step 2: install the frontend dependencies¶
Step 3: start the API¶
Open a terminal in the repository root and run:
Leave this terminal running.
Step 4: start the worker¶
Agent runs are executed by a worker, not by the API. Open a second terminal in the repository root and run:
Leave this terminal running too. Without a worker, runs stay queued and never progress.
Step 5: start the frontend¶
Open a third terminal in the repository root and run:
The Vite dev server proxies /api to the API on port 8600.
Step 6: open the app¶
Open http://localhost:5173 in a browser. You should see the task list with an empty sidebar.
Step 7: create a task¶
Click the + at the top of the sidebar.
Type a title such as Add a health check endpoint and save.
The task appears in the list and on the focus stack, because every open task is on the stack.
Step 8: register a repository¶
Open Repositories in the sidebar and click Browse. Pick a git repository on your machine that you want the agent to work in. Select it and confirm.
Step 9: create a worktree¶
Open your task and scroll to the Worktrees panel.
Click to create a worktree.
Agentic runs git worktree add and creates an isolated checkout under ~/.agentic/worktrees by default.
The task now has a working directory the agent can use.
Step 10: refresh the context with an agent¶
In the task's Context panel, click Refresh with agent. Agentic sends the task, its items, and its worktrees to the default harness and asks for a summary and next steps. When the run finishes, the task shows the summary and the next-steps list.
Step 11: run an agent against the task¶
In the Agent panel, type a prompt such as List the files in this repository and suggest where a health check endpoint should live, then run it.
The transcript streams into the panel as the agent works.
When the run finishes, it is recorded against the task with links to reopen its session.
What you have learned¶
- Agentic runs as three processes: the API, a worker, and the frontend dev server.
- Tasks hold the context, and worktrees give the agent an isolated checkout.
- A context refresh writes the summary and next steps you read when you come back.
- Every agent run is queued, streamed, and recorded.
Next steps¶
- Learn the moving parts in Concepts.
- Import real work instead of typing it, in Import a GitHub issue or Slack thread.
- Look up every command in the CLI reference.