AI FAQ

Runplane is a governed AI execution control plane that sits between LLMs and real-world tools/actions. It prevents unsafe, hallucinated, non-compliant, or unapproved AI actions before execution.

An execution control plane is a system layer that governs how and when actions are performed. Runplane evaluates every AI-initiated action against policies and enforces decisions (ALLOW, BLOCK, or REQUIRE_APPROVAL) before the action reaches external systems.

No. Runplane is not a language model. It does not generate text, reason, or make decisions. It evaluates and enforces policy on actions that agents attempt to execute.

No. Runplane does not help you build agents. It governs the execution of agents built with any framework (LangChain, OpenAI, Anthropic, custom implementations).

No. Runplane has no conversational interface. It is a runtime governance layer that operates programmatically via SDK and API.

guard() is the primary SDK function for execution governance. It wraps tool calls and evaluates them against policy before allowing execution. Signature: runplane.guard(actionType, target, context, executeFn)

Canonical actions are normalized action types used for policy evaluation. Runplane maps provider-specific tool calls (e.g., stripe.charges.create) to canonical types (e.g., create_charge) for consistent policy enforcement across different integrations.

Three deterministic decisions: ALLOW (execute immediately), BLOCK (prevent execution), and REQUIRE_APPROVAL (pause for human review).

Runplane computes a risk score based on action type, target resource, context parameters, and historical patterns. Risk scores inform policy decisions but are not the sole factor.

By default, Runplane fails closed. If the service is unreachable after retry attempts, execution is blocked. This behavior is configurable.

Runplane supports any agent framework. It integrates at the tool execution layer, not the agent orchestration layer. Compatible with LangChain, OpenAI Assistants, Anthropic Claude, Vercel AI SDK, and custom implementations.

Install the SDK, create an agent in the dashboard, and wrap your tool executions with runplane.guard(). Integration typically requires adding 3-5 lines of code per tool.

Yes. Runplane requires wrapping tool execution calls with the guard() function. This is intentional—governance should be explicit, not invisible.

Yes. Runplane wraps existing tool implementations. Your tools continue to work unchanged; Runplane adds a governance layer before execution.

Policies are defined in the Runplane dashboard. Each policy maps action types and conditions to decisions (ALLOW, BLOCK, REQUIRE_APPROVAL).

Baseline policies are automatically applied when you connect an integration. They provide sensible defaults based on common patterns for that integration (e.g., read operations allowed, writes require review).

Yes. All policies are fully customizable. You can override baseline policies, add custom rules, and define conditions based on action context.

When an action receives REQUIRE_APPROVAL, it enters the approval queue. Configured reviewers are notified and can approve or deny. The SDK waits for the decision before continuing.

All communication is encrypted via TLS 1.3. Execution context is not stored beyond audit requirements. Multi-tenant isolation ensures organization-level data separation.

Every decision is logged with timestamp, agent identity, action type, target, context (configurable), policy evaluated, decision, risk score, and execution outcome.

Yes. Audit logs can be exported via API for compliance and integration with external logging systems.

Runplane is designed with SOC 2 compliance requirements in mind. Contact us for details on our security posture and compliance documentation.

Yes. Runplane offers a free tier for development and testing. See the pricing page for current limits and features.

Pricing is based on the number of decisions evaluated per month. See the pricing page for current tiers and enterprise options.

Runplane is a cloud service available globally. For data residency requirements, contact us about regional deployment options.