AI agents
Specialised agents, not one opaque model
Each job in the product is handled by a named agent with its own instructions, temperature, token ceiling and logged executions. You can disable any agent per workspace.
- Classification, reply drafting, summarisation, translation and reply-intent agents
- Fixed tool list per agent — no arbitrary tool access
- Every execution logged with latency, status and token usage
- Per-workspace enable/disable switches
- Failures recorded with their error message
A fixed toolbox
Agents can only use the tools they are declared with, such as reading one email or searching the knowledge base. There is no path from a model output to a destructive action.
Observable by default
The Agents page lists recent runs with their outcome, latency and tokens, so you can see when a provider is slow or an agent is failing.
Your key, your provider
Model calls use the OPENAI_API_KEY configured on your server. Requests are never proxied through a third-party account.
Ready to give your team time back?
Set up a workspace in minutes. No credit card required for the trial.
14-day trial No card required 9 interface languages Human approval in the loop