Live · no signup · real policy engine
Watch AgentKey stop an AI agent.
AgentKey decides what an AI agent is allowed to do before the action happens, and creates evidence of the decision. Run a tool call below - the policy and the evidence chain are real.
AI agent
proposes a tool call
AgentKey
evaluates it against policy
Decision
ALLOW · DENY · approval required
Evidence
signed, chained, verifiable
refund_issue(amount: 199)
Runs the real policy engine and records a real signed evidence event. No signup, and no external tool is ever called.
No decision yet.
Pick a tool call on the left and evaluate it - the server returns the real policy decision and records signed evidence.
What just happened
Per-action authorization
The stored policy was evaluated for exactly this tool call, with this argument - not a blanket agent-level rule.
Default-deny
Anything the agent is not explicitly allowed to do is refused. The refund was denied because 199 exceeds the 100 limit.
Verifiable evidence
The decision was written as a signed event on a hash chain. The Verify button recomputes the hashes from the stored raw fields.
Now run it on your own agent
Zero dependencies, one line to wrap an existing agent. Everything you just ran - policy, decision, evidence - is the same code path your agent will use.
npm install agentkey-ai # JavaScript / TypeScript (Node 18+) pip install agentkey # Python (3.8+)
No account was needed for anything above. Creating your own agent and API key is the first step that needs one - keys are issued per agent under your account, so your policies and evidence stay isolated to you.