Organizations · Run

Put your whole team on it. Then watch it work.

Invite the people who do the work, give them agents built on your own documents, and see exactly who is using what. Selling one is optional.

3roles, scoped access
1library, every agent
0spreadsheets required

Bring your team in

Invite people. Decide what each one can touch.

Three roles, and they actually mean something. A developer can build and test all day without ever seeing what the org spends. It isn't hidden in the interface either. The numbers simply never come back for them.

OwnerEverything, including the plan, the wallet and who else gets in.
AdminPublish agents, invite people, and see what everything costs.
DeveloperBuild and test agents. Never sees spend or earnings.
Meridian Health · team2 / 4
DDana R.Owner
PPriya S.Admin
MMarco L.Developer
AAna K.Developer

invitations · single sign-on through WorkOS

Internal first

Most agents never leave the building.

That's fine. The marketplace is there if you want it, but the common case is an agent that knows your policies and answers your team.

Private by default

An agent stays inside your org until you choose otherwise. No listing, no pricing, no evaluation to pass. Just a tool your team can use this afternoon.

One library, every agent

Upload the handbook once and attach it to as many agents as you like. It counts against your storage a single time, and updating it updates everywhere.

Projects keep work apart

Give a project its own knowledge base and its own conversations, so the onboarding bot isn't reading the pricing deck.

We host it

Nobody on your team runs infrastructure.

This is the part teams usually underestimate. Retrieval, model routing, file storage and streaming are all ours to keep running. Nobody on your side is on call for it.

Nothing to deploy

No containers, no vector database to stand up, no keys to rotate. An agent is live the moment you save it.

Every model, one integration

Switch a team's agents from one provider to another in a dropdown. You never hold an API key.

Conversations survive

Someone's wifi drops mid-answer and it resumes where it stopped, even if they close the tab and come back later.

Customisable

Bring your own data. We'll attach it.

Your knowledge does not have to be re-uploaded to be useful. If it already lives in a database, a warehouse, a document store or a vector index you already run, we point an agent straight at it. You do not have to copy anything somewhere new.

Retrieval sits behind a single interface on our side, so swapping the store underneath is a settings change rather than a rebuild.

set up with our team, not a self-serve toggle. tell us about your stack

Sources we attach

Postgres

read-only replica

Snowflake

warehouse tables

SharePoint

document library

Confluence

spaces + pages

S3

bucket or prefix

Your vector store

existing embeddings

Your data stays where it is. We read it. We never take custody of it.

Track what happens

Who used what, and where it fell down.

Conversations, unique people, tokens and cost per agent, each of them comparable against the period before. The number that actually changes behaviour is the last one: what your agents get thumbs-down for.

Conversations

1,074+18%

People

63+9%

Tokens

3.2M+15%

Cost

412 cr+12%
AgentChatsPeopleUp-rateTrend
Policy Copilot6124184%
Onboarding Buddy2883491%
Incident Triage1741972%

What people complained about

Missed the question14
Wrong or outdated info9
Too slow5

cost and earnings render for owners and admins only · export any view as CSV

However you are staffed

You don't need an AI team to start.

The product is self-serve, but adoption rarely is. Two ways in, depending on who you have.

No engineers to spare

We are the team

You describe the work; we build, evaluate and ship the agents, then hand you a workspace your staff can actually run. Nobody on your side writes a prompt unless they want to.

  • We scope the first agents with you
  • We wire up your documents and systems
  • You get the finished workspace and the training
You already have engineers

We embed with yours

We sit alongside your developers for the first build, transfer the patterns, and step back. The agents stay yours, in your organisation, on your account.

  • Paired build on your real use case
  • Review of prompts, retrieval and evaluation
  • Handover, then support when you want it

Optional last step

And if one turns out to be good enough for other people?

Publish it. It has to clear the same evaluation as everyone else's, then it earns while your team keeps using it. Most agents never leave the building, and that is a perfectly good outcome too.

Shipping next · embed an agent in the product you already ship, while it stays managed and measured here.

01Internal usebilled at cost
02Clears evaluationsame bar as everyone
03Listed publiclyyour price, your page
04Earns70% of your markup

Free while you try it

Put your team on it this week.

Create an organization, invite the people who do the work, and point an agent at your own documents. Nothing to deploy and no card until someone is actually using it.

Talk to us

Tell us what your team spends its week on.

We'll tell you honestly whether agents help, which two or three are worth building first, and what it would take. No obligation to buy anything at the end of it.

Useful to include

01What your team does, and which bit is repetitive
02Where your documents and data live today
03Whether you have engineers to spare

a real person replies · usually within a working day