System 02 · AI Agents & Assistants
Agents that read every enquiry, recognise what’s being asked, use only the knowledge and tools you approve, and hand over to a person the moment it matters.
Demonstration with sample data. Decision steps are a readable summary of what the assistant did, not a model’s private reasoning.
Three kinds of assistant
Answers questions from your approved knowledge, qualifies enquiries, books calls and logs everything in your CRM.
Hands over sensitive or uncertain conversations with a summary.
Helps your people find answers in internal documents, draft replies and proposals, and summarise long threads.
Your team reviews and sends. The assistant drafts.
Carries out multi-step tasks across tools: reading a request, updating records, preparing documents and reporting back.
Runs within scoped permissions; anything irreversible needs approval.
Guardrails
An assistant is only useful if you can trust it. These controls are part of every build, not optional extras.
Assistants answer from knowledge you’ve reviewed. If the answer isn’t there, they say so and hand over.
Each agent can only read and change what its job needs, nothing more.
Pricing, complaints, legal questions or low confidence route straight to a named person.
Transcripts, actions and hand-offs are recorded so you can review and improve them.
We surface what the assistant did (which sources, which tools, why it escalated) as plain steps your team can audit.
Tell us which conversations take the most time. We’ll design an assistant around them, with your team in control.
