Your key, your model.
The platform calls the AI provider you configure: a cloud service or a model on your own hardware. No third party by default.
A SELF-LEARNING AI YOU CONTROL
It drafts. It cites. It learns. It never acts alone.
The assistant answers customers from your knowledge base, helps agents draft replies, triages new work and proposes new articles from resolved conversations. You bring the key. Humans approve everything.
FOUR PRINCIPLES
The rules the AI can never break.
Grounded answers only.
It answers strictly from the brand's published articles, retrieves the most relevant chunks, cites them, and returns a confidence score.
Humans approve everything.
Drafts are inserted for the agent to edit, never auto-sent. Learned articles wait in review until a person approves.
Transparent and budgeted.
Every answer shows its confidence. Every call is metered against a monthly budget you set.
AUTO-ANSWER
First-line support that never sleeps.
First-line support that never sleeps.
And never guesses.
Visitors get instant answers from your own help articles, sources cited. When the AI is unsure, a human takes over with the full conversation in hand. Every answer it sends is badged, so you can check its work any time.
Citations are intersected with the brand's own sources. A made-up reference can never attach.
AGENT COPILOT
Three one-click helpers wherever agents write.
One click gets you a draft, a summary or a plan. Not quite right? Ask for "shorter" or "more formal", or just say it by voice. Whatever it writes, only a human hits send.
Draft reply
A suggested response, built from the conversation and your knowledge, inserted for the agent to edit.
Summarise
The story so far, in brief. Catch up on a long thread in one read.
Suggest next steps
A proposed plan for the case, ready to act on or discard.
AUTO-TRIAGE
A second pair of eyes on everything new.
When new items arrive, the AI reads them and suggests priority, type and where they belong, as a chip on the item. Informational by design: no field changes without a person doing it.
THE SELF-LEARNING LOOP
It learns from every case you close.
It learns from every case you close.
You approve what it publishes.
Resolved conversations turn into drafted help articles that wait for a person's yes. Published articles are cited in the very next answer, and Knowledge gaps ranks what your KB cannot answer yet.
AI drafts are never auto-sent. Learned articles never publish without human approval.
BRING YOUR OWN KEY
Your provider, your budget, your data.
Plug in your key and the AI runs on the provider you choose, even a fully local model, so nothing has to leave your infrastructure. Keys are encrypted on save and never shown again. You set one monthly cap and watch spend live, by capability, model and day.
Anthropic
OpenAI
OpenAI-compatible
Groq
Ollama (local)
HARD TO FOOL
Built for hostile input.
Questions, not instructions.
Whatever a customer types is treated as a question, never as instructions. The AI cannot be talked into ignoring its rules.
Blind to other brands.
It was never shown another brand's data, so it cannot reveal it.
Failure has a fallback.
Provider failure never means silence: the visitor is handed to a human and a ticket is created automatically.
The cap comes first.
Budget caps run before any billed call.