Connect your agents
Give your agents structure. Keep the record.
For CTOs, AI engineers and platform leads: describe your domain once - what your entities are, how they relate, what rules bind them. Every agent you connect gets that same picture, scoped to the session and current at the moment it runs, instead of whatever a prompt happened to carry. Keep your stack, your models, your product. Each new agent inherits the same limits, routing and record as the last one.
The problem
Every agent you add reasons over your data slightly differently.
Your agent works in the demo, and you can't explain to Legal why it won't do something unexpected once it's live. Every agent you connect interprets your data slightly differently, because nothing forces a shared definition of what a field means. Every new customer wants their own constraints encoded, and today that means writing bespoke context by hand, per integration, instead of once.
How it works
Your agents plan01 against the picture your model defines, decide02 inside the limits it enforces, and act03 onto the same record as everything else.
01
Describe it once
Entity types, relationships and rules, written down once. Every agent that connects inherits them instead of working your domain out again per integration.
02
Scoped to the session
Each session gets what's current and what's relevant, plus the specific set of actions it may take. Not whatever the prompt happened to carry.
03
Same limits for every agent
Every agent action inherits the limits, routing and record the rest of the platform already enforces. Governance isn't a separate product.
Connect through whichever interface fits. What the agent gets on the other side is the same every time. MCP and the schema-export interfaces are part of the Enterprise tier - what that involves.
Data integrations
- Read and write through a REST API
- Connect agents over MCP (early access)
- Forward operational email straight in
- Import spreadsheets and bulk files
- Get a webhook when state changes
- Define your schema once, in one place
Accountability
You keep the product. We hold the record.
You own
- The product your customers see.
- Your choice of LLM. Bring your own.
- Your data, in your workspace.
The AI owns
- Handing each session the right slice of context.
- Holding the limits you set, on every call.
- Putting every action on one record.
Under the hood
The structure isn't new infrastructure.
Your Model already types every entity and its constraints; Resources hold the live state, versioned as it changes. That is what runs in our live pilot today. The MCP server, in early access, puts a standard interface on top of it.
What this is not
- Is the MCP server available today?
- It's in early access, not general availability. The Model and Resources underneath it are what runs in our live pilot today. Create a workspace and request access from there.
- Do we have to migrate off our own agent stack?
- No. Metronome sits on top of your existing agents and models. You keep your stack, your models, your product surface.
- Is this a memory layer or a vector database alternative?
- Neither. A memory system returns what is similar. Metronome returns what is true, permitted and owned, right now.
Point one agent at it.
Describe one slice of your domain, connect an agent you already run, and see what it gets on the other side: current context, scoped to the session, with the actions it is allowed to take already attached.
Metronome is in beta. You create a workspace and define your first Model there - nothing to install, and you keep your own agents and models.