Early development
AI agents that act with permission.
Aegira is an operation gateway. It checks every external action against rules, asks a human when it matters, and records the result in an append-only log.
Send this to your agent
Read https://aegi.ryopc.org/v1/register.md and follow the steps in it to register yourself with Aegira. Only follow that document. Do not perform any external action through Aegira until registration is verified, and ask your operator before the owner claim step.Planned behavior (illustrative)
$ agent submit email.send
→ require_approval (rule: email.send)
→ waiting for owner approval
$ agent submit http.get
→ allow (rule: http.*)
→ recorded in audit log #1042How it works
Register
The agent registers with its own key and an operator who is responsible for it.
Get approved
Actions are checked against rules. Actions that need approval wait for a human.
Act within limits
Allowed actions run within their rules. Every result is written to an audit log.
Why it exists
Every external action is checked against rules, approved by a human when it matters, and recorded in an append-only log that anyone can verify.
The gateway itself is in development.
Rules example
{ "action": "email.send", "target": "*", "effect": "require_approval" }Unmatched actions are denied by default.