What is Ananke?
Ananke is a governance runtime that sits between agents and tools, enforcing policy, approval, auditability and recoverable outcomes.
Open source / governed execution
Project Ananke
An AI governance runtime between agents and tools.
Last reviewed 31 August 2026
Active development
MIT-licensed public repository.
What problem does it solve?
Agents can call tools that change files, send messages, deploy software or expose information. A simple success or failure result does not provide enough control or evidence around those actions.
How does it work?
Ananke wraps governed tool calls in structured outcome envelopes. Read-only actions can be allowed quickly, while risky writes can be bound to human approval and recorded in an audit trail.
Key properties
- MCP-compatible and protocol-agnostic execution boundary.
- Policy and approval before governed actions.
- Typed, recoverable outcomes instead of opaque success/failure.
- Audit records for the governed path.
Important boundary
Ananke only governs what is routed through its chokepoint. It does not make an ungoverned path safe by existing elsewhere.
Source and evidence
Inspect the governance boundary.
- View source repositoryPublic MIT repository
- Read architecture and design notesVerified docs directory
- Read the security policyVerified file
Related projects
The Fates
Umbrella runtime architecture.
Runtime Contracts
Shared contract vocabulary.
FAQ
Is Ananke an MCP server?
No. MCP connects tools; Ananke adds a governance chokepoint around calls reached through MCP and other transports.
Does Ananke make every tool call safe?
No. Governance depends on routing the relevant action through the enforced path and on the policy and approval rules applied there.