Runledge

Let AI agents work in production.
On your terms.

Stripe is infrastructure for payments. Runledge is infrastructure for agent actions. From coding agents updating production to claims agents issuing payouts—set the rules, route sensitive decisions to human approvers, and keep a verifiable audit trail.

Start free See the quickstart ↓
Live checkillustrative

Tool call · claude-code

db.migration.apply

target: production

Your rule

Production writes need 2 admins

Decision

Held for approval

1 of 2 signed

Receipt

Ed25519 signed · identity and time

Your agent: Proposes

OpenAI, Anthropic, LangGraph or your own loop.

Runledge: Checks

Allowed, blocked, or held for approval.

Your code: Acts

Runs only what was allowed.

Runledge: Records

Signed receipt, cost by department.

Coding agents: review production changes.

Check migrations, infrastructure changes and deploys before they run, with production exceptions held for your chosen engineers.

Insurance: review claim exceptions.

Route payout and policy-change exceptions to your adjusters and keep their decisions ready for review.

Explore industry workflows

One HTTP call to your first policy check.

Check proposed actions through REST or OAuth MCP while keeping your models, agents and tools.

  1. 01

    Create a workspace

    Create a free account and workspace.

  2. 02

    Write one rule

    Require a manager’s approval for refunds over $500, then create an API key.

  3. 03

    Ask before acting

    Call Runledge before your agent’s tool runs, follow its decision, and record the outcome.

Make your first policy check.

Configure a workspace rule, then send a sample proposal to check it before your code runs.

Terminal
curl --fail-with-body https://runledge.lovable.app/api/public/v1/preflight \
  -H "Authorization: Bearer $RUNLEDGE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"action":{
    "agent":"billing-agent",
    "action_type":"stripe.refund.create",
    "target_system":"Stripe",
    "metadata":{"amount_usd":250}
  }}'
SDK examples, MCP setup and in-VPC deployment

The obvious question

“Why can’t the agent check its own actions?”

A process can’t be its own security boundary. If the agent is compromised or rewritten, checks inside it disappear with it.

Lives inside the agent

Keep policy outside agent code, with your executor enforcing the decision or the broker holding provider credentials.

One person can change it

Version policy changes separately and require your chosen reviewers for risky action exceptions.

Nothing to show an auditor

Give auditors signed identity-and-time receipts and file-hash manifests they can verify independently.

Rewritten for every agent

Share workspace rules across the models, frameworks and agents your teams connect.

Models will be wrong sometimes.

Your execution layer can't be. Runledge sits between the tool call an agent proposes and the system it would change.

  1. Agent plane

    Your models and workflow engine propose tool calls for policy review.

    Components
    • Temporal, Inngest or your own loop
    • OpenAI
    • Anthropic
    • Open-weight models

  2. Runledge runtime

    Check proposals against your rules and collect the approvals needed before execution.

    Components
    • Policy check against enabled rules
    • Approval quorum you define per rule
    • Optional credential broker for bound provider hosts
    • Signed identity and time receipt

  3. Systems of record

    Your executor runs allowed actions, or the credential broker calls the provider host bound to its stored key.

    Components
    • Stripe
    • Salesforce
    • NetSuite
    • Workday
    • Internal APIs

  4. Security and audit

    Your collector pulls ordered audit events, and auditors verify export integrity against a signed manifest.

    Components
    • Splunk
    • Datadog
    • Elastic
    • Signed audit bundles

Sign up to review the architecture in portal docs →

For every team that signs off

Built for the teams accountable for production.

Dry-run policy changes against real history.

Compare draft policies against recent action records before applying a versioned revision.

Choose who approves. Export signed evidence.

Give auditors read-only access to signed receipts, export manifests and public verification keys.

Attribute agent cost and value by department.

Compare reported model spend, estimated labor value and risk exposure by department, with CSV and PDF exports.

Find your workflow

Where this goes

Start inside your company. Extend trust beyond it.

Today: Your agents, your rules

Check proposed actions, require approval for exceptions and keep signed records of reported outcomes.

Preview: Across companies

When your agent acts at a partner, a signed ticket proves your rules allowed that exact action. The partner still applies its own.

Direction: Evidence others can rely on

Receipts that partners, auditors and, one day, insurers can check for themselves.

Your first check takes minutes.
Your last incident, hopefully, never.

The Developer plan is open now. No waitlist, no card. Sign up, get an API key and make your first check.

Start free

Enterprise

Need a security review, self-hosted deployment or a custom contract? Tell us where to reach you.