Welcome to Orbitali
Developer infrastructure for deploying low-latency voice AI agents.
Orbitali is the production-grade developer platform and core runtime infrastructure for building, deploying, and operating real-time AI voice agents. By providing a low-latency telephony integration and a unified voice runtime, Orbitali handles the heavy lifting of speech processing and orchestration, leaving developers to focus entirely on their business logic.
Why Orbitali?
Traditional voice AI platforms chain three separate systems together:
Speech-to-Text (STT) → LLM Reasoning → Text-to-Speech (TTS)
(~300ms) (~400ms) (~200ms)
+ network latency between services = 900ms - 1400ms TTFBThis latency breaks natural human conversation patterns.
Orbitali is built exclusively on Gemini Live (via Vertex AI). Speech recognition, reasoning, and voice synthesis happen in a single model in one pass:
Gemini Live (Unified STT + LLM + TTS)
= 300ms - 500ms TTFB (Natural Human Response Latency)This design choice prioritizes performance over personalization. Developers trade the ability to swap models or voice engines for a conversational agent that stays in the natural human response zone.
Core Capabilities
1. Inbound Voice Telephony
Orbitali uses a Bring Your Own Carrier (BYOC) model. You lease and maintain your inbound phone numbers directly in your Twilio or Telnyx accounts, then configure them to route calls to Orbitali. You pay the carriers directly for connection minutes without any markup, while Orbitali charges a flat runtime fee of 0.10€ per minute (billed in 30-second increments).
[!NOTE] Orbitali is strictly optimized for inbound call handling (e.g., reception triage, client intake, and appointment booking). Outbound cold-calling campaigns or automated mass dialing are not supported.
2. Built-in Knowledge Base (RAG)
You can upload company policies, product sheets, or FAQ documents in PDF or Markdown format directly. Orbitali automatically chunks your files, embeds them using text-embedding-004, and queries them locally using pgvector inside PostgreSQL. Your agent uses the native search_knowledge tool to answer questions instantly, without you needing to build search APIs.
3. Custom Tools & Webhooks
For database lookups or transactional flows (e.g., checking appointment availability, booking a table, looking up an invoice), your agent triggers custom tools. Orbitali executes these by POSTing agent:tool-call events to your configured Server URL webhook, letting you connect any backend, CRM, or database securely.
4. Ephemeral Browser Calling
If you need voice experiences directly in your mobile app or website, Orbitali supports direct browser calling. Your backend generates an ephemeral session token, and your frontend connects directly to the voice runtime via WebSockets to stream microphone audio.
Platform Responsibilities
| What Orbitali Handles | What Your Backend Handles |
|---|---|
| Bidirectional telephony media streaming | Business logic and private data |
| Real-time ASR, LLM reasoning, and TTS synthesis | Webhook endpoints for custom tool execution |
| Active listening, barge-in, and interruption cut-offs | Ephemeral session token generation |
| Local document chunking, indexing, and vector search | Dynamic prompt construction (optional) |
| Call history records, tool execution logs, and usage tracking | Post-call automation (CRMs, scheduling systems) |
Next Steps
- Quickstart: Deploy your first voice agent using the dashboard or API.
- Concepts: Understand the architecture of agents, prompts, tools, and sessions.
- Guides: Dive deep into webhook signature verification, tool design, and browser integrations.
- API Reference: Programmatically configure and scale your agents.