Execution Control Platform

How Runplane wraps tool execution and enforces policies at runtime.

AI agents execute real-world actions through tools:

  • Database modifications
  • External communications
  • Infrastructure provisioning
  • Financial operations

Runplane sits between your agent and these tools. Every tool execution is wrapped with guard(), which evaluates the action against your policies and enforces a decision: ALLOW, BLOCK, or REQUIRE APPROVAL.

How Runplane Works

The integration model is SDK-first. You bring your existing tools, Runplane derives canonical actions automatically from your tool definitions, and the SDK wraps execution.

1. Bring Your Tools

Existing agent tools from any framework

2. Auto-Mapped Actions

Canonical types generated automatically

3. Baseline Policies

Ready-to-use, fully customizable

4. Wrap with guard()

Enforced at runtime

AI Agent calls tool
runplane.guard()
Policy Evaluation
ALLOWREQUIRE_APPROVALBLOCK
Tool executes (or is stopped)

Runplane works with any agent framework. Your AI system continues to use LangChain, Vercel AI SDK, CrewAI, or custom implementations — Runplane wraps the tool execution layer.

SDK Integration

The primary integration is through the SDK. Wrap any tool execution with guard().

Wrap tool execution
await runplane.guard(
  "delete_employee_record",
  "hr_system",
  { employeeId: "emp_7421" },
  async () => {
    await deleteEmployeeRecord("emp_7421")
  }
)

How guard() works:

  • 1. Sends action type + context to Runplane
  • 2. Receives decision: ALLOW, BLOCK, or REQUIRE_APPROVAL
  • 3. If ALLOW: callback executes immediately
  • 4. If BLOCK: callback never runs
  • 5. If REQUIRE_APPROVAL: waits for human approval

Policy Engine

When you import tools, Runplane applies baseline policies that provide ready-to-use governance. You remain in full control — customize any policy to match your requirements.

Policy Configuration

Policies can target:

  • Action type — delete, deploy, payment, send_email
  • Target — production database, external API, payment gateway
  • Context — amount thresholds, time windows, agent identity

Decision Outcomes

ALLOW

Callback executes immediately. Low risk, policy permits.

REQUIRE_APPROVAL

Paused. Human review required before execution.

BLOCK

Callback never runs. Policy violation.

Human Approval Workflows

When an action requires human approval, guard() holds execution until an authorized operator reviews and approves the request.

Approval Process

1

Agent calls guard()

AI agent attempts to execute a tool wrapped with guard()

2

Runplane holds execution

Policy returns REQUIRE_APPROVAL, callback is paused

3

Operator reviews request

Dashboard shows action details, context, and risk assessment

4

Callback proceeds or is blocked

If approved, callback executes. If denied, callback is blocked.

Audit Trail

Every action that passes through guard() creates an immutable audit event. This provides complete visibility into what your AI systems attempted to do and what decisions were made.

Recorded Attributes

agentId— Which agent made the request
actionType— Canonical action type
target— Target system
decision— ALLOW / BLOCK / REQUIRE_APPROVAL
timestamp— When the request was made
context— Action context for policy evaluation

Importing Tools

Use the dashboard to import your existing tools. Runplane automatically maps them to canonical action types and applies baseline policies. You stay in control — customize any policy to match your requirements.

Supported import sources:

  • LangChain tool definitions
  • Vercel AI SDK tools
  • OpenAPI specifications
  • Manual tool registration