See how it behaves, inside the process you already run.

It fits inside your stack.

Every change Agent Etna proposes lands as a real GitHub PR labelled agent-etna. Your branch protection, CODEOWNERS, required reviewers and CI run on it exactly the way they run on every other PR. There is no proprietary runtime, no rewrite, and nothing to migrate to. Your model API and your repository stay where they are.

The review and CI your team already trusts is the most valuable part of the system. Agent Etna's job is to feed it a steady stream of ready-to-review PRs, not to route around it.

Every change comes with numbers.

Every proposed change ships with a pass rate, a regression count, and side-by-side outputs against the previous version — measured against your agent's own behavior, not a generic benchmark.

• Baselines come from your agent's own system prompt, not a handwritten spec.
• New behavior is checked against real traffic in the shadow, so drift shows up before your users see it.
• Cost and latency are tracked at the percentile, not just the average — averages hide the tail.

One workspace, not one login each.

A workspace holds the agents, not a person’s account. Everyone on it sees the same dashboards, opens the same traces, and reads the same scenario a teammate flagged — a review is a conversation on one record, not a screenshot pasted into Slack. Billing is pooled across the workspace rather than charged per person per run, so nobody has to decide whether a simulation is worth expensing.

When someone leaves, you remove the seat. The agents, the traces and the regression corpus stay where they are.

Everything the UI does is an API call.

Opening a PR, reading its checks, kicking off a change — every action in the dashboard has a plain HTTP endpoint behind it, and the same surface is exposed to other agents over MCP. If you would rather drive Agent Etna from your own code than from the dashboard, that path is first-class — see what it looks like for a programmatic caller.