Research & Education · Free educational license
One consistent view of every agent in your multi-agent system — across frameworks, across hosts, including the ones you didn't write. On the wire, payload-blind, with nothing instrumented.
A measurement instrument, not a platform
Maya is a per-agent network telemetry and governance layer for multi-agent systems. It identifies each agent from the host and observes its traffic payload-blind — never terminating TLS, never decrypting — with no SDK, no code change, and no framework instrumentation. One declaration line per agent, and Maya emits a clean, structured record of what that agent reaches, who it delegates to, and where it strays from what you declared — exportable as OTLP or JSON, straight into your own notebooks and backends.
Measure
Agentic systems
Get ground truth for multi-agent runs: per-agent flows, provider calls, message rates, and the agent-to-agent delegation graph across hosts — for LangGraph, CrewAI, AutoGen, or your own agents, one consistent stream, none of them instrumented.
Study
Agent safety & control
Declare an agent's authorized envelope, then watch Maya flag when observed behavior contradicts it — the basis for studying misbehavior, containment from outside the agent, and agents that hold a valid credential yet act out of role.
Teach
Courses & labs
Stand up a reproducible multi-agent environment students can deploy and reason about — delegation graphs, deviations, and containment as hands-on, repeatable exercises — without asking every student to instrument every agent.
What makes it different
Never terminates TLS, never decrypts. Maya reads envelope metadata — who, where, how much — not your content. The right tool when the object of study is structure, scale, or containment, not message semantics.
Identity is derived from the host, per agent — finer than a pod or a service, with no cooperation from the agent itself. It sees the black-box agents you didn't write and can't instrument.
Maya watches the agents you name and nothing else. No guessing, no silent surveillance of unrelated workloads on the host.
Structured deviation telemetry is the product. Route it out to your own judge — a notebook, an LLM, Open Policy Agent, or your SIEM — and decide what it means.
How it works
You deploy Maya in an environment you control — your data never leaves it. Two packages: a host-side package that runs on each agent host, and the Warp appliance that governs their traffic, provisioned after a brief host-inventory discovery so there's no version-matrix guesswork.
Step 01
A short call to map your host inventory and what you want to study or teach.
Step 02
We hand you the two packages, recommended compute specs, and a runbook.
Step 03
Add one line to each agent's existing deployment config to name it. No code changes.
Step 04
Read per-agent telemetry, map delegation graphs, flag deviations, plug in your own judge.
Where Maya is today
Per-agent observation, delegation-graph mapping, and declared-vs-observed deviation telemetry are what we lead with. Programmable enforcement is available and evolving.
We're direct about what is proven versus in progress — and a design partnership is exactly where that conversation belongs.
Does Maya fit your setup?
Design partners, not customers
We partner with a limited number of research labs, universities, and educational programs each term. You get the full substrate under a free educational license, a step-by-step runbook, and hands-on onboarding — because we want to learn from how you push it. In return: candid feedback, a chance to learn together from real multi-agent deployments, and case studies or citations only if and how you're comfortable.
Maya — Research Edition · Educational license, free of cost