How do you know your agents are behaving?
It is the question every executive eventually asks about an AI deployment, and the one most teams cannot answer with anything sturdier than a Slack thread. Multiple teams, multiple agents, one answer for engineering, compliance, and leadership — backed by signed evidence rather than assertion.
One organisation, many agents, one view.
A central dashboard shows every agent's health, every team's activity, every change in flight. Project-level scopes keep teams autonomous; the central view sees without intervening.
- Per-team workspaces with isolated settings
- Role-based access (admin, engineer, viewer)
- Per-agent quotas and budget caps
- Cross-org search: "every agent that uses GPT-4"
The audit log is already there.
Every change to every agent is a pull request on your repository, so the record of what changed, who approved it and what it was replayed against lives in your own git history and survives Etna being offline. Encryption at rest, access controls and change management are how the system is built, not a checklist on top of it. SOC 2 is not started; our controls are documented and we share them on request.
GDPR-aligned data handling. PII stripping before any data is retained. A breach-incident register that is part of the product, not a Notion page. Any user can export or erase what is stored about them from their account.
Self-host or hosted.
No vendor lock-in. If your agents handle regulated data and have to stay inside your VPC, Agent Etna runs there — deployed against your own Postgres, with your model keys never leaving your perimeter. If you would rather not operate it, the hosted version handles uptime, backups, and upgrades. The product is the same in either case.
A self-hosted Etna talks to a GitHub Enterprise Server behind your firewall end to end: sign-in, repository reads and writes, PRs, git clones into the sandbox, and GitHub App installation. One environment variable does it — GITHUB_SERVER_URL=https://ghe.internal.corp — and the API root is derived from it, so there is no half-configured state where authentication still reaches the public internet. GitLab Self-Managed and Bitbucket Server are pointed at the same way, with GITLAB_API_BASE and BITBUCKET_API_BASE. Nothing about your repositories leaves your perimeter.
One piece is not there yet, and we would rather you hear it from us than from your security review. Execution sandboxes run in E2B’s cloud on an API key (your own key, if you bring one) rather than inside your AWS or GCP account. It is on the roadmap, ahead of most of it, for the first customer who needs it; if that is you, say so in the first conversation and we will scope it into the pilot.
Pilot in two weeks.
Write to us. We will scope a focused pilot on one of your agents, set up the workspace, walk your engineering team through it, and you decide. No procurement marathon. No six-month sales cycle.
Talk to us about a pilot.
Book a 15-minute slot or send us a message — both on the contact page. Reply within 24 hours.