Platform
Flows, AI agents, models and rules in one place, inside the systems you already use.
A customer writes in Zendesk. The flow starts.
A run
Seven steps, always in the same order, whatever the work is.
A system, a channel or a schedule starts the work.
The steps, the conditions and who does what.
The knowledge and the data that belong to the question.
The model that fits, within your policy and budget.
AI agents do the work with real tools.
Back in your system, or in an app.
Who, when, which model and what it cost.
AI teams
AI agents work in teams with a lead, members and reporting lines. A task lands with the lead, who splits it and hands each piece to the right member. Every agent has its own model, its own instructions and exactly the tools you gave it.
More about AI agentsRefund question on order #20417
Iris
Support Lead
·
Milo
Support Engineer
·
Nova
Support Agent
·
Otis
Billing Specialist
·
Work ledger
Each member claims its own piece, so one instruction is split only once.
Routing
Each run picks a model within your policy and records why. A light question goes to an efficient model, a failing provider hands over to a fallback model, and your region stays your region. The keys and the invoice stay with your provider.
More about routingmistral-small-latest·Light question, efficient modelEfficient
A short chat turn, so it ran on the provider’s efficient model.
Pal
Pal is the assistant every person in your organisation gets. Ask what waits for you, how a flow is doing or what last week cost. When a question turns into work, Pal hands it to the AI agent that can do it.
Meet PalPal
Only you can see this chat
What’s waiting on me today?
Developers
A REST API with an OpenAPI specification, an MCP server, your workspace as code and signed webhooks. Whatever you do this way follows the same approvals, budgets and audit trail as the app.
For developersversion: 1
governance:
require_tool_approval: true
monthly_budget_usd: 400
teams:
- slug: support
monthly_budget_usd: 250
agents:
- name: Iris
rank: lead
model: claude-sonnet-5
fallback:
- { provider: mistral, model: mistral-medium-latest }
tools: [zendesk, mollie]A workspace described as code: a budget, an approval rule, a team, an AI agent with its model, fallback and tools.
Control
Anything that changes something in your systems waits for a person, who sees the exact call. Budgets are checked before every run and stop at the cap. Every run, decision and change lands in the audit trail. All of it is on from the first run, for flows, chat, the apps, the API and MCP.
More about controlNeeds your decisionjust now
Refund €48.00 for order #20417
Creating a refund in Mollie always needs approval before it runs.
Try it: the call only runs when you say so.
From the product
Also built in
AI agents answer from your documents, your website and your connected help centre, with the source beside the answer. With strict grounding an agent answers only from those sources, or says that it cannot.
Save real prompts as eval cases and run them with the agent’s actual tools. A small change to an instruction can then never quietly break what already works.
You can also start from a ready-made team. See the blueprints
We start by listening to your question. Then we show you a demo that fits your organisation.