Matimo™ Gateway is the one endpoint your AI agents talk to instead of talking to model providers directly. Every call is identified, checked against your policies, routed, and recorded, with no changes to your agent code and with your own provider keys.
Not yet released. Availability, packaging, and pricing will be announced here first.
# before
client = OpenAI(
base_url="https://api.openai.com/v1"
)
# after: governed
client = OpenAI(
base_url="https://<your-gateway>/v1"
)One line changes. Your agent logic, prompts, and framework stay exactly as they are.
Most teams ship agents faster than they can govern them. Provider keys end up scattered across repos and laptops, nobody can say which agent made which call, and the audit trail is whatever the provider's dashboard happens to show. Matimo Gateway puts one controlled doorway in front of every model call.
Works with OpenAI-compatible and Anthropic-compatible clients. Change the endpoint and you are governed. No rewrite, no migration project, no framework lock-in.
Every call is tied to a registered agent identity, so "who made this call" has a real answer, for your security team and for your auditors.
Your rules are applied before a request reaches the model. Stopping a bad call beats explaining it later.
Bring your own provider keys. They stay on the platform side, so they never sit in an agent's process, a laptop, or a notebook.
Send traffic to the providers you choose from a single control point, instead of hard-wiring a provider into every agent.
A single, consistent record of agent activity across every framework and team, ready for the questions compliance will ask.
Suspend or revoke an agent and its next call through Gateway is refused. No redeploy, no hunting for the key, no waiting for a code change to ship.
That guarantee covers calls that pass through Gateway. It is why the endpoint matters: anything that goes around it is, by definition, ungoverned.
The Agent Governance Development Kit is the open-source Python SDK that pairs with Gateway. It brings agents built in LangChain, Google ADK, CrewAI, AutoGen, or plain Python under governance, and adds what a proxy alone cannot. MIT licensed, Python 3.13+, available on PyPI today.
AGDK is built to work with Matimo Gateway, which is coming soon. Suspension is rapid, not instant: the next call that reaches Gateway is denied immediately, while the SDK's own local check is polled on a heartbeat and can lag by up to one interval. If the platform is unreachable, tool checks fail closed by default.
from matimo_agdk import Governor
governor = Governor.from_env()
governor.start()
with governor.run("my-run"):
result = governor.guard(
my_tool_fn, name="search"
)(query="roaiq")Aim your agents' model traffic at Gateway. For deeper control, add Matimo AGDK and get identity, telemetry, and per-tool-call checks too.
Your policies decide what each agent may do. Calls are identified, checked, and routed to the provider you choose, using your keys.
Every call leaves a record. When security, compliance, or a customer asks what your agents did, you have the answer.
Let every team keep its framework while you keep one control point for model access, keys, and spend visibility.
Agent identity, policy enforcement, and an audit trail without relying on every developer to wire it in correctly.
Ship agents without carrying provider keys or building governance plumbing. Change an endpoint and get on with the product.
Tell us what you are running and we will reach out as early access opens. In the meantime, Matimo OSS and Matimo AGDK are available today.