AI agents and chatbots that do real work.
We build AI assistants that connect to your own systems, so they can do more than answer questions. They can book appointments, look up records, process requests and pass difficult cases to a person. We build on open-weight models by default, and on a frontier model where the task calls for it. The chatbot on this site is one of ours, and it runs on an open-weight model.
Get in touchTry it: the replies are scripted, the interface is real
Most chatbots in use today look answers up in a fixed knowledge base. They can tell you things, but they cannot do anything. An agent can use tools such as APIs, databases, calendars and forms, and it chooses which to use as the conversation goes on.
That means an agent can handle a task from start to finish. It can book a call when a visitor asks. It can find lower-carbon options for a buyer and rank them. It can sort an incoming support request and pass it to a person when it cannot resolve it.
Every agent we deliver comes with a set of tests: example inputs and the outputs we expect. The tests run with each release, so you can see how the agent behaves and whether a change has broken anything.
Frequently asked questions
What's the difference between a chatbot and an agent?
A chatbot answers questions from a fixed knowledge base. An agent also has tools, so it can search, book, retrieve information, process requests and hand over to a person, depending on the situation. In practice, a chatbot tells the user what to do and an agent can do it for them.
Which models do you build on?
We start with open-weight models, which are generally free to use commercially and can run on your own infrastructure. We also build on frontier models such as Claude and GPT where the task calls for it, although they usually cost more. Which one we recommend depends on where your data has to stay, how quickly responses are needed, and cost.
How long does a deployment take?
A production-ready agent with two to four integrations usually takes four to eight weeks from brief to launch, including testing and handover. More complex projects, such as several agents or retrieval over private data, take eight to twelve weeks.
Do you hand over the code?
Yes. All the code goes into a repository you own, running on infrastructure you control. We do not build on a platform that only we can operate.