Quick Start

This walks you through your first agent, end to end: add a repo, open a branch, and let an agent work while you watch its status and spend. It takes about two minutes.

Prefer to call Lattis from your own code instead of using the IDE? Jump to Point your client at it.

1. Connect a model

Open Lattis. In Settings, either:

• Connect a cloud provider — sign in to Anthropic or OpenAI, or paste an API key (see Cloud Providers & Accounts), or
• Connect a local model server — LM Studio, Ollama, llama.cpp, vLLM, or any OpenAI-compatible endpoint you run yourself (see Local Models).

Your agents will run against whatever you connect here — no per-agent setup.

2. Add a project

In the sidebar, open Projects → Add project and pick a local git repository. The repo and its branches appear in the sidebar.

Lattis ships with Sawyer, its built-in agent — the default, with nothing to install, so you can start right away. It also drives the coding-agent CLIs you already have — claude, codex, pi, or opencode — install one and it shows up in the agent picker too. (Prefer plain chat? Try Conversations against any connected model.)

3. Open a branch

Click New branch, give it a name, and Lattis will:

1. Create the branch and a dedicated worktree (a separate checkout on disk),
2. Launch your default agent there, in its own terminal,
3. Open the branch’s tiled workspace around it — Changes, Files, Context, and Spend, ready in their own panes.

Type a task into the agent and let it work. Re-opening the same branch later reattaches to the running session instead of starting a duplicate.

4. Track it while it runs

As the agent works, watch:

• The stoplight next to the branch — green while it’s working, amber when it’s waiting on you, red if it errors. See Stoplights & Status.
• Changes — the files and commits it’s producing, with line counts.
• Spend — the running dollar cost, per model. See Changes & Spend.

5. Ship it

When the work looks good, use the branch’s actions to Open Pull Request, then watch its CI checks and merge — all without leaving Lattis. See Pull Requests & CI.

Lattis then offers to Request Review and ranks the people who wrote the code you changed. When a teammate asks you for a review, it lands in Pending Reviews in the sidebar, and one click opens the branch as a guided diff you can comment on — with an optional agent pass to find issues first. See Code Review.

That’s the loop: open a branch, let an agent work, answer it when it needs you, review, merge. Open several at once and you’re running a team.

Point your own client at it

Everything above runs on a local endpoint that speaks both the OpenAI and Anthropic APIs, on 127.0.0.1:5288. You can point any compatible client or SDK at it directly:

curl http://127.0.0.1:5288/v1/chat/completions \
  -H 'Content-Type: application/json' \
  -d '{
    "model": "qwen3-4b-instruct-2507",
    "messages": [{"role": "user", "content": "Hello!"}]
  }'

See the Public API for the full surface, including the Anthropic-compatible endpoints and streaming.