AgentLedger
Sign inStart free
The control plane for consequential AI actions

Give your AI agents authority—with boundaries.

AgentLedger evaluates high-impact actions before they execute. Allow safe work, block policy violations, and route sensitive decisions to a person—with a complete audit trail.

Protect your first agent View integration
OpenAI, Claude, Gemini, or custom Free for 1,000 monthly evaluations
LIVE POLICY EVALUATION
AGENTRefund Copilot
online
Proposed actionIssue a $400.00 refundpayment_intent: pi_3U…
Policyrefunds_over_250
DECISIONHuman approval required

Execution is paused. No Stripe call has occurred.

Evidence recorded evt_7d90
One policy boundary across every model and tool
OpenAIAnthropicGeminiStripeCustom agents
Deterministic governance

Three decisions. No ambiguity.

Your model proposes the action. AgentLedger decides whether it is authorized before your executor runs it.

01

Allow

Known, low-risk actions continue immediately within the agent’s delegated authority.

decision: "allow"
02

Require approval

Sensitive actions wait in a durable queue until an authorized person decides.

decision: "require_approval"
03

Block

Actions outside policy stop before they can reach the connected system.

decision: "block"
Built for accountability

Know what acted, why it acted, and who approved it.

AgentLedger creates a defensible record around every consequential decision without taking custody of your model credentials.

Scoped agent keysEvery key is bound to an agent and its permitted actions.
Signed completion webhooksVerify approval, denial, and execution outcomes in your application.
Immutable evidenceRequests, decisions, approvers, timestamps, and execution results stay connected.
Provider neutralYour OpenAI, Anthropic, or Gemini credentials remain in your own application.
Incident-ready evidence
  1. 10:42:08.114Action requestedrefund-copilot · issue_refund · $400
  2. 10:42:08.192Policy matchedHigh-value refund approval
  3. 10:44:31.608Approved by managerIdentity and decision retained
  4. 10:44:32.041Stripe execution succeededRefund ID and response recorded
Signed outcome delivered 204
One call before execution

Connect an agent in minutes.

AgentLedger sits between agent intent and tool execution. Keep your existing stack, model, and executor.

JavaScriptPOST /api/evaluate
const decision = await fetch("https://app.agentledgersecurity.com/api/evaluate", { method: "POST", headers: {"x-agentledger-key": process.env.AGENTLEDGER_API_KEY }, body: JSON.stringify({ agent: "refund-copilot", action: "issue_refund", amount: 40000}) } ).then(response => response.json());
response{ "decision": "require_approval", "approval_id": "apv_…" }
Simple pricing

Start free. Scale when agents do more.

Free

$0 / month

For prototypes and your first protected agent.

  • 1,000 policy evaluations
  • Approval workflows
  • Signed completion webhooks
  • 30-day audit access
Start free
PRODUCTION
Pro

$99 / month

For teams governing production agent actions.

  • 50,000 policy evaluations
  • Unlimited protected agents
  • Signed webhooks with retries
  • CSV audit exports
Start with Free
YOUR AGENTS CAN MOVE FAST

Your controls should move faster.

Put a policy boundary around your first consequential AI action today.

Create your free workspace