1. AI on this website
- Blog posts: drafts are written with help from Claude, an AI model made by Anthropic, on topics we choose. An engineer checks every draft against the official documentation, corrects it and approves it before anything is published. Nothing is published automatically.
- Comments: every blog comment is read and approved or rejected by a person. No automated decisions are made about you.
- The live workforce page: shows the status of automated AI agents doing engineering work as they plan, build, test and deploy. Dimmed figures are placeholders that keep the scene company when it's quiet, and the page says so.
- There's no chatbot or AI assistant on this site, and every enquiry is read by an engineer.
No visitor's personal data is sent to an AI service by this website.
2. Our approach to the AI systems we build
We build AI voice agents and automation, including on LiveKit, Kamailio and FreeSWITCH, as part of dependable telephony, not as a replacement for it. We favour the least open-ended method that solves the caller's problem: fixed audio for a fixed message, deterministic routing for a known decision, and a conversational AI agent only where a genuine conversation adds value.
3. Safeguards we design in
- A clear route to a person or another safe service, wherever the journey matters.
- Explicit no-input, no-match, timeout and error outcomes, so a failure ends somewhere sensible.
- Approved instructions, knowledge and tools only, with permissions limited to the task.
- Confirmation, or a deterministic check, before anything sensitive or irreversible.
- Testing before go-live, and call records and reporting so behaviour can be reviewed.
4. Providers and your data
AI systems usually rely on specialist providers for speech recognition, speech generation and language models, which can be hosted services or self-hosted models. We choose them with you for each project, based on latency, cost, language and where your data may be processed. The providers, and how they handle data, are agreed with you before any of your data is processed by them.
5. Telling callers and limitations
You're responsible for any disclosure the law, a contract or reasonable caller expectation requires, including telling callers when they're speaking to an automated agent or being recorded.
AI output can be inaccurate, incomplete, biased or delayed, and speech recognition varies with accent, language, codec and background noise. AI must not be treated as emergency service access, an identity check, or professional legal, medical, financial or safety advice.
6. Questions and concerns
Email: [email protected]
Phone: +44 161 820 1210
Also see: Privacy notice · Cookie notice · Terms