Custom AI assistants
Assistants that answer from your own documents and cite the page they used. Agents that take approved actions in your systems and hand over to a person when they should.
Support assistant
Connected to 214 help center pages
Can a customer on an annual plan get a refund after 30 days?
Yes, within 60 days of purchase. Annual plans are refunded pro rata for the unused months. After 60 days the plan runs to the end of its term and simply doesn't renew.
Refund policy, section 2
Billing FAQ, updated March
What should the agent tell them to do?
Send them to Settings, then Billing, then Request refund. The refund reaches their card in 5 to 10 business days.
Billing help, refunds
What we build with AI
- Assistants over your documents
- Help-centre answers, internal policy lookups and sales enablement, grounded in your own files with a link to every source.
- Agents that take approved actions
- Create a ticket, update a CRM record or draft a refund, with limits on what the agent may do alone and a person to approve the rest.
- AI features inside your product
- Summaries, search, classification and drafting built into the app your customers already use, not a separate chatbot.
- Search across scattered data
- One question across your wiki, drive, tickets and database, with permissions respected so people only see what they could already open.
Evaluation run120 real questions from your team
Ships The candidate passes every check, so it replaces the current model.
How we keep AI reliable
- An evaluation set before launch
- We collect real questions from your team and the right answers, then score every version of the assistant against them. A change ships only if the score holds.
- Sources on every answer
- Each answer links to the document it came from, so your team can check any answer in one click.
- Cost you can see
- We log the cost of every answer and set budgets with alerts, so spending stays where you planned it.
- Models you can swap
- Claude, GPT and open-weight models sit behind one switch point, called a model router. When a better model arrives, we test it on the same questions and switch only if it wins.
We also build iOS and Android apps and web apps and automation. Most projects use more than one. See how a project grows.
- v1 An assistant over your documents
- v2 Approved actions in your systems
- v3 The assistant inside your app
- v4 The next model, on the same tests
Common questions
- How long does it take to build a custom AI assistant?
- A first assistant over a defined set of documents is usually scoped as a release of a few weeks. We confirm the timeline and a fixed quote after a scoping call.
- Will our data be used to train AI models?
- We use the business APIs of providers such as Anthropic and OpenAI, which by default do not train their models on data sent through those APIs. Where your data must stay in your own cloud, we can use open-weight models you host.
- Which AI model is best for our use case?
- It depends on the task, so we test several. We compare accuracy on your evaluation set, response time and cost per answer, then choose, and we keep the choice easy to change.
- Can you add AI to our existing app or website?
- Yes. We work in your codebase and infrastructure, and usually start with one feature your users ask for every day. See our mobile app and web app work for how that fits together.
- What happens after launch?
- We monitor accuracy and cost, add new questions to the evaluation set as they come in, and upgrade models when a better one passes the same tests.
What are you building?
Tell us in a few lines. We reply within two business days with questions or a time to talk, and the first call is free.
Describe your project