Agentic IDE
Lattis includes a built-in agentic IDE for running multiple coding agents side by side. Instead of juggling separate tools and terminals, you manage them from one place — all pointed at the models you’ve connected in Lattis.
Supported agents
Section titled “Supported agents”| Agent | Description |
|---|---|
| Codex | OpenAI’s coding agent. |
| Claude | Anthropic’s coding agent. |
| Pi | Lattis’s own agent. |
Run them individually or in parallel — for example, having two agents tackle the same task so you can compare results, or splitting work across several at once. The default agent (used when you click a branch) is configurable in Settings.
Projects, branches, and worktrees
Section titled “Projects, branches, and worktrees”The Projects screen organizes work by git repository. Add a project (a repo), and its tasks are simply its git worktrees — Lattis lists them live rather than tracking task state separately:
- Open a branch to launch the configured agent in that worktree, alongside a persistent side shell. Re-opening the same branch reattaches to the running session instead of starting a duplicate.
- Create a new branch to spin up a fresh worktree and start an agent there — ideal for working on several tasks at once without them stepping on each other.
- Each open worktree can host multiple agent tabs, so you can run, say, Codex and Claude on the same task and compare.
Branch metrics
Section titled “Branch metrics”Alongside each worktree, a metrics panel tracks two views of the branch:
- Changes — the commits ahead of the base branch (
main/master) and the uncommitted files in the worktree, each with their line churn (+/-). - Spend — the usage and cost attributed to that branch, broken down per model.
Driven against your models
Section titled “Driven against your models”Each agent runs against the models Lattis already serves, so the same local models and cloud providers you’ve connected are available to your agents — no separate configuration. Lattis points each agent’s API client at its gateway automatically (Claude and Codex via their base-URL settings; Pi talks to the daemon directly).
Because every agent’s traffic flows through Lattis, its work shows up in your usage and cost breakdown alongside everything else, so you can see exactly what each agent is spending.
Why it matters
Section titled “Why it matters”- One place for every agent — no per-tool setup or context switching.
- Mix vendors freely — pair the right agent with the right model for the job.
- Accountable — agent spend is attributed and tracked like any other usage.
See Cloud Providers to connect the models your agents will use, and Usage & Cost to track what they spend.