An orchestration layer for B2B contact data. You keep the provider accounts, the contracts and the raw cost. We run everything on top of them — and expose it to your agents properly.
There is no endpoint to call, no key to generate and no documentation, because there is nothing yet to document. We are not going to publish coverage figures, latency numbers or a price we have not tested.
What does exist is the underlying work. We have been sourcing, enriching and verifying B2B contact data as a delivered service since 2022, and running agent tooling in production on our own systems. This is that capability, packaged — not a new business we are guessing at.
Available todayIf you need this solved now, we already do it as a managed service. That part is real and available today.
The established way to buy contact data is to hand the whole problem to one vendor. They hold accounts with a dozen data providers, run your query across all of them until something comes back, and bill you a credit. It works, and for many teams it is the right answer.
It also means the margin, the vendor relationships and the switching cost all sit with them rather than with you. You pay retail for data that was bought wholesale, and the day you want to leave, the contracts are not yours to take.
We are building the other half of that: the orchestration, without the resale. You keep your own accounts with whichever providers you choose. We run the waterfall across them, handle the caching, the deduplication and the verification, and expose the whole thing to your systems — and to your AI agents — through one interface.
The usual model
What we are building
You hold the accounts with each data provider and pay them their rate directly. We never touch the billing relationship, take a cut of it, or become the party you have to negotiate with. If you want to swap a provider out, that is a configuration change and not a migration.
Most tools grew an AI interface after the fact, wrapping an API that was designed for a human clicking a button. We are starting from the agent: tools with narrow contracts, explicit failure modes, and enough context for a model to decide correctly whether to call them at all.
An agent that enriches every record it sees will empty your credit balance by lunchtime. Teaching a model when not to call a tool is the harder half of the problem, and it is one we have already had to solve on our own systems.
Every lookup records which provider answered, what it cost, whether the result was cached and why the waterfall stopped where it did. A bill you cannot reconstruct is a bill you cannot argue with.
Tools an AI assistant can call directly, designed so a model can decide correctly whether to call them at all.
How the tools are designed →Plain HTTP underneath all of it, because an agent-only product cannot be put in a cron job.
Planned endpoints →No figures yet — but the shape is decided, and the shape is the argument.
The model →We would rather hear that the design is flawed now than after we have built it. If you run enrichment at any volume, tell us which providers you use, what your monthly spend looks like, and what you would want an agent to be able to do with it.
Early access shapes what gets built first, and we will be honest about timelines rather than promising a quarter we cannot commit to.