Agents that answer, act and ask first.
Chat and voice agents that answer from your knowledge, act only through approved tools, and wait for the customer before anything changes.
Everything an agent is made of.
No developers needed to build one. Click through the parts.
Pick the model per agent.
Run local models through Ollama, vLLM or any OpenAI-compatible server, or use cloud models. Provider keys stay on the server.
- Optional fallback model and per-agent tuning.
Answers from your own material.
Files up to 25 MB, pasted text and web pages, indexed by a local embedding model. Access is set per source.
- Grounded only, with citations required when you choose.
Act only through approved tools.
Call an API through a connector, or on the Suite run a whole workflow. Tools are denied by default and each message has a call budget.
- Try it from the tool screen before an agent ever uses it.
Checks on every message, in and out.
Input limits, prompt-injection blocking and personal-data filters before the model. Prompt protection and a personal-data scan on every reply.
One agent, every channel.
Studio chat, your websites and apps through access keys, and voice. The same knowledge, tools and audit everywhere.
- Access keys with rate limits, allowed origins, expiry and retention.
Checked by the server, not promised by a prompt.
Tool inputs are validated on the real value before your system is called, and only the customer can release an action.
- “Yes, but make it £200” does not count. Only a button press or a clear, short yes.
- The model cannot release an action on its own.
The same agent picks up the phone.
One agent serves chat and voice with the same knowledge, tools, checks and audit. The on-premises engine keeps audio inside your network.
- Recording notice before the greeting.
- Reads back details before any action, and hangs up on silence.
Everything on record, nothing exposed.
Workspaces
Shared knowledge, hand-offs between agents, tool allow-lists, allowed web domains, budgets and default guardrails.
Masked history
Every channel is masked. Revealing a value needs permission and is written to the audit log first.
Retention per key
Up to ten years or no limit, storage warnings at 75, 90 and 100 percent, legal hold and recorded deletion.
Bring a real customer question.
We will show an agent answering it from your material, every check it runs, and where it would live on your servers.