Curizen Autonomous Voice Agents
Voice agents that finish what your customer started.
One platform to build, configure and deploy autonomous voice agents that search your own knowledge, act on your own systems, and take the request all the way to done.
- Grounded in your own knowledge
- Voice and chat in one session
- Versioned, promotable releases
Field Service Agent
release 0.0.7 · production
Agent activity
Intent
Damage assessment → booking
Knowledge
3 passages · hybrid search
check_availability
200 · 240 ms
book_appointment
awaiting customer confirmation
Every turn, retrieval and system call is logged against the release that produced it.
8 languages
One agent, primary language plus presets.
Hybrid retrieval
Semantic and keyword search over your own content.
Two environments
One Testing and one Production release per agent.
AES-256-GCM
Every credential encrypted at rest.
The platform
A complete platform, not a single agent.
Everything it takes to build, ship and run autonomous voice agents at production scale - in one place, owned by your team.
Build voice agents
Identity, role, personality and guardrails as structured settings, with an AI-drafted system prompt you edit section by section.
Configure without code
Dials, rules and forms instead of scripts. Describe a REST endpoint once and it becomes a tool the agent can call mid-conversation.
Deploy multiple agents
Run a fleet, each with its own knowledge, tools and voice. Immutable releases move through Testing and Production on your say-so.
Manage conversations
Every session kept with its transcript, the passages it cited and the systems it called - per agent, per release.
Monitor analytics
Call volume, success rate and latency for every tool, with the stage each request reached. Nothing happens off the record.
Integrate your stack
REST APIs, MCP servers, your own database, WhatsApp, scheduling backends and model providers - credentials encrypted at rest.
Product tour
From a blank draft to a live agent, in one sitting.
Watch a real agent get built - guardrails, knowledge base, a booking tool wired to a live endpoint, and the release that puts it in front of customers.
curizen - building an autonomous voice agent
- 01
Define
Role, goal and audience, plus what the agent must always do, never do and escalate.
- 02
Ground
Upload documents, paste text or point it at a page - indexed for hybrid retrieval automatically.
- 03
Connect
Describe your endpoints once; they become typed tools with guardrails around them.
- 04
Release
Freeze the agent into a numbered version, test it, then promote it to production.
Capabilities
What the agent can actually do.
Holding a conversation is the easy part. These are the capabilities that decide whether a customer leaves with the job done.
Natural, interruptible voice
Pick the voice and tune stability, speed and similarity per agent. It knows when it is speaking and when it is listening, so people can cut in.
One agent, eight languages
English, German, French, Dutch, Spanish, Italian, Polish and Arabic - identity, prompt and tools carry across every one of them.
Voice and chat, one session
A visitor can start by typing and switch to speaking without repeating themselves. Same conversation, same transcript.
Answers from your own content
Documents, text and web pages split into single-idea passages, then searched semantically and by keyword at once - with the source behind every answer.
Actions on your own systems
Typed tools with validation, timeouts, retries and fallbacks, so the agent books the slot instead of describing how to book it.
Escalation that keeps context
Explicit rules for when a person takes over, approval gates before sensitive actions, and the full transcript waiting for whoever picks it up.
Where teams put it to work
Booking & rescheduling
Checks real availability and confirms the slot inside the same conversation.
Service & repair intake
Asks the questions a trained advisor would, then routes the case correctly first time.
Policy & coverage answers
Grounded in the document that governs the answer, worded consistently every time.
Status & handover
Looks up where things stand, and hands over with full context when a person is needed.
Enterprise ready
Autonomy, with a paper trail.
An agent that acts on your systems has to be answerable for it. This is how the platform is built, not a roadmap.
Encrypted at rest
Every credential stored with AES-256-GCM, decrypted only for the moment a call is made and never returned to a browser.
Your data, your database
Knowledge bases can live in your own project, isolated per customer rather than pooled behind a shared tenant column.
Immutable releases
A committed version can never be edited. Changing what customers talk to means promoting a new release, deliberately.
A complete audit trail
Every tool call, knowledge-store operation and conversation logged against the release that produced it.
FAQ
The questions that come up first.
What can the agent actually do beyond talking?
It searches your own knowledge base for the passage that answers a question, and it calls your systems through tools you define - checking availability, creating a booking, looking up a record. Each tool has its own validation, timeout, retry and fallback behaviour, and can require human approval before it runs.
How do we stop it from going off-script?
Behaviour is configured, not hoped for. Every agent has explicit rules for what it must always do, must never do, must ask before doing, and when it must escalate to a person - plus policies such as confirmation before sensitive actions and sensitive-data masking.
How does a change reach production?
Edits accumulate in a draft that is never live by itself. When you are ready you commit a numbered release that freezes the whole agent: configuration, prompt version and the exact version of every tool it owns. That release runs in Testing until you promote it, and each agent holds exactly one Testing and one Production release at a time.
Can the same agent handle text as well as voice?
Yes, and in the same session. A visitor can start by typing, switch to speaking mid-conversation, and keep one continuous transcript. The embeddable widget connects in text immediately and only asks for the microphone when the visitor chooses to talk.
Where is our data stored?
An agent's knowledge base can be pointed at your own database project, keeping content, embeddings and audit history in your account rather than a shared store. Credentials for it are encrypted at rest and are never sent back to a browser after they are saved.
Put your first agent on a real conversation.
Ninety minutes with your documents and one of your systems is usually enough to tell whether an autonomous agent belongs in your customer journey. Let's find out on your content.
We map one real journey end to end - the questions, the systems, the outcome that counts as done.
We build the agent live on your own documents, so you hear it answer from your content, not a script.
You leave with a working agent in Testing and a clear view of what production would take.