Fence
Stops one agent's change from landing over another's without a check.
Aevos, Inc.
Every consequential AI action needs a receipt.
When two or more AI coding agents work in one repository, one agent's change can overwrite another's, and a change can land with no record of who made it or why. Aevos is settlement for that case: a fence, an admission check and a receipt for every landing, so you can inspect who changed what, under what authority, and why.
Write to hello@aevos.io. We respond within 2 business days.
Who it is for
Aevos is for engineering teams that run two or more AI coding agents against the same git repository. When agents share a codebase, one agent's change can overwrite another's. We call that a stomp. The earlier work can then disappear from the code without anyone noticing.
If your team runs a single agent, or each agent works in its own repository, the problem Aevos addresses is less likely to apply.
What Aevos does
Aevos is a fence, admission and receipt layer. It helps agents' changes land safely and leaves a record you can inspect.
Admission
Decides whether a change is allowed to land, under the authority it was given.
Receipt
Records who changed what, under what authority, and why, for each landing.
Aevos works under MCP and agent harness hooks, on git.
Two ways to work with us
Both offerings are fixed-fee professional services, invoiced in US dollars.
Stomp Audit
From $3,000
A fixed-fee audit of your multi-agent coding workflow for collisions and overwrites ('stomps') and for unsafe landings. You receive a written report with findings and fixes.
Typical delivery: 1–3 weeks.
Landing Safety Pilot
From $9,000
A scoped pilot that installs the Aevos settlement layer on one repository for your team, with a receipt for every agent landing.
What the audit can and cannot tell you
- The Stomp Audit reads the git history it is given and flags overwritten lines it can detect in the scanned history, for human review.
- Flags are review aids. They are not a measured accuracy or recall figure, and they do not prove that any behavior was preserved.
- No finding does not mean nothing was lost. The audit covers the history it reads, and shallow clones hide older history.
- Each flagged candidate is triaged in the written report. None is discarded.
We state these limits up front so you can judge what the audit will and will not show.
Start with an email
Tell us which offering you are considering, roughly how many AI coding agents work on the repository, and which git host you use. We respond within 2 business days.
Read pricing and payment terms first, if you prefer.