Cryptographic permission tokens that control exactly what your AI agents can do. Define, verify, revoke — all in real time, all in your control.
Three steps from zero to controlled AI agents
Tell us what your agent does in plain English. Our LLM parses it into a structured permission set automatically.
We generate a cryptographic permission token with your agent's exact capabilities encoded and verifiable. Immutable, tamper-proof, yours.
Your agent's runtime checks permissions before every tool call. Unauthorized actions never execute. One API call, sub-50ms.
Every feature designed around one principle: agents should only do what you explicitly allow.
Permissions are cryptographically signed and tamper-proof. Not stored in a config file your agent can rewrite. Verified every time.
Revoke a token and the agent loses all access immediately. No propagation delay, no cache to expire, no trust to revoke.
Permissions checked both at the token layer AND at the API endpoint. Both must match. An attacker would need to compromise both systems.
Describe permissions in plain English. "Read files and search the web but never run shell commands." We handle the rest.
Tokens expire automatically. 1-year default with 6-month, perpetuity, or custom options. No stale access hanging around.
Agent identifiers use hash-based pseudonyms by default. Your agent's real identity stays private while permissions stay verifiable.
Add AgentLeash to any agent framework in minutes. Works with OpenClaw, LangChain, CrewAI, AutoGPT, or your own runtime.
// Verify before every tool call const resp = await fetch("https://api.tokenform.com/api/v1/verify", { method: "POST", headers: { "Authorization": `Bearer ${API_KEY}`, "Content-Type": "application/json", }, body: JSON.stringify({ agentId: "my-research-agent", toolName: "web_search", }), }); const { allowed, reason } = await resp.json(); if (!allowed) { console.log(`Blocked: ${reason}`); // → "Tool 'web_search' not in allowlist" }
Beta coming soon. Join the waitlist to get early access.