AI Governance to manage prompts and context for your AI Agents.

A2A, REST API and Git-like workflows. Every request logged; lineage and audit built in.

Goal

Governance that doesn't slow shipping. Protocol-first (A2A, MCP), versioned prompts/context, and a living agent registry.

Open source

Run it yourself, extend the API, or contribute. Low-key, practical, protocol-first.

Integrate your way

A2A

Sandarb is an AI Agent for AI Governance. It can communicate with other AI Agents over the A2A protocol. Other agents call GET /api/a2a for discovery, then POST /api/a2a with a skill (get_context, validate_context, get_lineage, register).

a2a-protocol.org →

REST API

All resources and actions over HTTP. Inject context from your agent: GET /api/inject?name=my-context. List and manage via /api/contexts, /api/agents, /api/organizations.

Git-like

Propose edits with a commit message; approve or reject in the UI. Contexts and prompts get versioned history. Push-based: register agents via ping or manifest; Sandarb tracks approvals and revisions like a lightweight PR flow.

What we solve

Single place for approved prompts and context; agents pull via API or A2A.
Audit trail: who requested what, when (lineage for compliance and incidents).
Manifest-based agent registration; git-like versioning for prompts/context.
Sandarb runs as an A2A agent so others call it for validation and approved context.