Authorization layer for AI agents

Give agents authority without giving them control.

AgentLedger evaluates, approves, executes, and audits consequential AI-agent actions.

Open dashboardSign in

Live demo results

Every action audited
100%
Duplicate charges in replay test
0
Always wins over allow
Deny
Built on
  • Supabase
  • Stripe test mode
  • OpenAI
  • Jev
  • MCP
  • Next.js
  • Postgres RLS
  • Realtime
For merchants

Make your store agent-ready.

Verify your domain, publish an authoritative catalog, and make your products purchasable by MCP agents through buyer-controlled policy.

Explore merchant access

One feed.
Any MCP agent.

Unverified websites still work through a fallback that always requires human approval.

Product

Authorization

Deterministic policy

Every intent runs through the same rules engine — budgets, merchants, categories and scopes. Same input, same decision, every time.

Approvals

Human approval in real time

High-impact actions pause and wait for a person. Approvals land live in the dashboard and resolve the agent the moment you decide.

Audit

Tamper-evident audit

Each decision, execution and receipt is hash-chained into an append-only ledger you can verify end to end.

Guardrails

Stop the bad action before it costs anything.

  • Jev signals

    Prompt-injection and price-anomaly signals feed straight into the decision.

  • Trust ≥ 95

    Merchants below the trust threshold are escalated or denied.

  • Kill switch

    One toggle halts every agent action instantly. Deny always wins.

Verified Merchant Registry

Know who your agent is paying.

Merchants prove control of their domain. Agents check the registry before checkout, and anyone can look up a public verification page.

/verified/<domain>
MCP

Plug any agent in over MCP.

Agents connect with OAuth consent, then submit intents through the MCP server. Every call is scoped to the delegation you granted.

→ submit_intent
  merchant  "store.example"
  amount    $42.00
← decision  REQUIRE_HUMAN
  reason    "above auto-approve limit"
How it works

Every consequential action gets a principal, scope, policy, decision and receipt.

  1. 01
    Principal
  2. 02
    Delegation
  3. 03
    Intent
  4. 04
    Policy
  5. 05
    Approval
  6. 06
    Execution
  7. 07
    Receipt
  8. 08
    Audit
DenyAuto-approveRequire human

Ship agents you can actually trust.

Open dashboard