Talk with our team, and be first to hear when new capabilities ship.
Book a walkthroughEvery workspace on the Growth and Enterprise plans already has it. An owner flips it on under Workspace settings, picks which frameworks it may read, and the assistant appears in the side rail for anyone with reviewer permissions. Nothing to install and no separate contract.
A pair of hosted general-purpose models handle drafting and summarisation, and a smaller in-house classifier we trained on de-identified control text does the routing. We publish the current line-up, and the date it last changed, in the trust centre.
Only the snippet needed to answer the request in front of you, and only under a zero-retention agreement — prompts are discarded by the provider once the response is returned. Nothing you send is used to train anyone's model, ours included, and you can turn off outbound calls entirely.
Requests run inside the same tenancy boundary as the rest of your workspace, encrypted in transit and at rest, scoped by the role you already hold. Every prompt and response is written to the audit log, so a reviewer can reconstruct exactly what the assistant saw.
Across 4,180 graded questionnaire responses last quarter, 87.4% were accepted by the reviewer without an edit. The assistant cites the control or policy it drew from on every answer, and it declines rather than guesses when the evidence is thin.
Our internal governance programme is built on ISO 42001 and the published risk-management framework for AI systems, and it inherits the access, logging and vendor-review controls from our existing SOC 2 and ISO 27001 scope. The mapping is available to customers under NDA.