LA CLÉ FOR CHATGPT, CODEX, AND AI AGENTS
Give agents authority. Keep control.
LA CLÉ lets compatible AI agents work under a bounded mandate: explicit allowed and blocked actions, limited data access, a budget, an expiry date, and human approval when required. Secrets stay out of the conversation.
A complete control loop
Before, during, and after an agent action
Define
Set the mission, permissions, prohibited actions, data, budget, expiry, and approval mode.
Approve
A mandate created by an agent remains inactive until its owner reviews and activates it.
Verify
Stage exact actions for review, record the real provider outcome, and revoke authority at any time.
TRY THESE PROMPTS
Three concrete workflows
“Create a seven-day mandate that lets my agent prepare calendar events but requires my approval before creation.”
“Check whether this €300 booking fits a €400 budget, then ask me before authorizing it.”
“Stage this email under my mandate, but do not send it until I approve the exact recipients and body.”
11 FOCUSED TOOLS
From bounded delegation to revocation.
The tools draft and save mandates, simulate a policy decision, verify public proof, list limits and pending approvals, stage supported Gmail or Google Calendar actions, expose approved actions to an executor, record real outcomes, and revoke a mandate.
WHY LA CLÉ
Control that lives outside the conversation
- Deterministic limits checked on the server.
- Configurable human approval for sensitive actions.
- Immediate revocation and no plaintext mandate-token storage.
- Public proof separated from private action authorization.
Essential limitation
LA CLÉ governs only agents and services that call its API or tools. It cannot intercept an unconnected AI, and it does not replace the security, consent, or payment controls of Gmail, Google Calendar, a merchant, or a bank.