Pricing
Start free for 14 days. Limits are enforced in the product, so what you see here is what you get.
Limits at a glance
Billing questions
No. The instruction builder is a structured form and the flow is drag-and-drop. If you do want code, there is a prompt editor, a REST API with scoped keys and webhooks for everything.
The LLM layer is a provider abstraction. OpenAI, Anthropic and any OpenAI-compatible gateway (Azure OpenAI, OpenRouter, vLLM, Ollama) work out of the box by changing environment variables. Voice uses ElevenLabs Conversational AI.
Browser voice works immediately. Telephony is behind an adapter interface: connect Twilio or another carrier with your own account, and inbound, outbound and campaign calling use the same agent configuration.
Retrieved knowledge takes priority over model knowledge, the prompt requires a "no reliable answer" response when nothing matches, answers can be forced to cite their source, and every tool result is recorded so you can audit exactly what the agent did.
Every organisation-scoped query goes through a tenancy helper that appends the organisation filter, PostgreSQL row-level security policies enforce the same rule in the database, and a test suite specifically attempts cross-tenant access on each resource.
Encrypted with AES-256-GCM using a key that never leaves the server. Credentials are never returned to the browser — the UI only ever sees the last four characters — and they are redacted from every log and audit record.
Yes. Docker Compose brings up PostgreSQL, Redis, the app and a background worker. Point DATABASE_URL at Supabase or your own PostgreSQL; nothing in the code depends on a managed vendor.
The action is refused with a clear message naming the limit, and the dashboard shows usage against every quota before you get there. Limits are enforced whether or not billing is switched on.