The agent asks
It runs alicit run with a profile, a time limit and a justification. Nothing else changes in its workflow.
Private beta Credential approvals for AI agents
Your agents ask. You decide.
alicit puts your decision in front of every credential an agent uses. You see the exact operation on your iPhone, and Face ID signs your answer.
How it works
One command. One decision.
The agent does not hold a standing token. It asks for the smallest profile it needs, for a short time, and it says why.
Your iPhone shows the request
You see the operation, its exact inputs and the reason. Swipe right to approve. Swipe left to decline.
The credential stays sealed
The command gets a local proxy, not the real token. When the command stops, alicit revokes the token.
Approve
Read it. Then sign it.
Each request is one card. The card shows what will happen, not a fingerprint that you must trust.
$ alicit run --profile github-issues --ttl github-issues=5m \ --justification "List open issues for the release review" \ -- gh issue list --repo alicit-ai/alicit › Waiting for approval on iPhone… › Approved with Face ID. Token expires in 5m. › Command finished. Token revoked.
Waiting · 1
List issues in alicit-ai/alicit
“List open issues for the release review”
DeclineApprove
Face ID signs each decision
Principles
Small, short and signed.
The smallest profile
An agent asks for one server-owned profile, such as read access to one repository. It does not get a broad key that works everywhere.
The shortest time
Tokens last minutes, not months. Each time the agent needs a new credential, it asks again.
A key only your phone holds
The approval key lives in the Secure Enclave of your iPhone. The server checks the signature before it issues anything.
Built on OpenBao
alicit runs in front of an OpenBao vault that you control. Policy decides what an agent can request. You decide what it gets.
Let your agents work. Keep the keys.
alicit is in private beta. Ask us for an invitation.