Maya is a security fabric for AI agent traffic — purpose-built to identify, classify, and govern the new class of network flows that legacy infrastructure cannot see. This is where the developer docs will live.
Maya inspects flows generated by AI agent processes — including the application-layer protocols agents actually speak — and exposes structured telemetry, identity, and policy controls through a clean operator API.
It runs alongside your existing security stack. It does not require modifying your agents. It does not add inline latency to agent calls.
// mayagentic.io — coming soon ├── overview // what Maya is, what it isn't ├── architecture // data / control / management planes ├── agent-identity // how Maya identifies agent processes ├── traffic-classification ├── policy-model // writing rules in agent terms ├── telemetry // OpenTelemetry-format export ├── integrations │ ├── mcp-servers │ ├── openai-compatible │ └── grpc-services ├── deployment │ ├── vpc-appliance │ ├── on-prem │ └── air-gapped └── sdk ├── python └── node
1. Decide in the kernel; classify off the path. Pre-provisioned agent blocks drop at wire speed, zero-leak. Everything else passes and is classified in parallel — the agent is never stalled. 2. Identify by what the agent is, not its IP. A dozen agents behind one shared IP are a dozen distinct identities, derived from the agent's runtime context — not its address. 3. Speak the language agents actually use. Not just IP and ports. The protocols on top. 4. Express policy in terms operators can reason about. Capabilities, not CIDR ranges. 5. A human approves every AI-generated change. Always. No exceptions. No countdown timers.
We're opening early access to a small group of teams running agents in production.
→ [email protected]