Platform

Persistent memory
for voice AI agents

Two API calls. Sub-200ms latency. Your voice agents remember every caller, every conversation, every detail — across every call.

<200ms

Retrieval latency

Voice-grade speed

99%

Retrieval accuracy

Hybrid search + re-ranking

2

API calls to integrate

Sync + Context

0

Infrastructure to manage

Fully managed API

Core Capabilities

Built for voice AI from the ground up

Every feature designed for the unique requirements of real-time voice conversations.

Intelligent fact extraction

Extracts atomic facts from call transcripts. Names, preferences, issues, and commitments — filler and greetings automatically discarded.

Contradiction detection

When a caller updates their info ("actually, my new address is..."), memory updates intelligently. Always current, never cluttered.

Hybrid search

Semantic + keyword matching with re-ranking for 99% retrieval accuracy. Finds the right memories even with vague queries.

Per-caller isolation

Every phone number or caller ID gets a completely isolated memory space. No data leaks between callers. Ever.

Voice-grade latency

Sub-200ms retrieval designed for real-time voice. Memory loads before the AI even starts speaking. No awkward pauses.

Multi-format ingestion

Send call transcripts, JSON messages, audio files, or PDFs. Automatic transcription and fact extraction from any format.

Real-time webhooks

Get notified when memories change. Trigger CRM updates, analytics events, or follow-up workflows automatically.

Data portability

Export all memories as JSON or CSV anytime. Your data, no lock-in, full portability. GDPR-friendly by design.

Integration

Two endpoints. Five minutes to integrate.

Works with Vapi, Retell AI, Bland AI, Twilio, ElevenLabs, or any custom voice stack. No infrastructure to provision.

POST/api/sync

After call → Save memories

When a call ends, send the transcript. Orthanc extracts facts (name, preferences, issues, commitments), deduplicates, and stores. Returns immediately.

// After call ends — sync the transcript
await fetch('/api/sync', {
  method: 'POST',
  headers: { 'Authorization': 'Bearer sk_...' },
  body: JSON.stringify({
    userId: callerPhone, // +1234567890
    messages: callTranscript
  })
})
POST/api/context

Before call → Retrieve context

When a call starts, fetch the caller's memories. Inject into the AI system prompt. Your voice agent now has full context. Sub-200ms.

// Incoming call — get caller context
const { memories } = await fetch('/api/context', {
  method: 'POST',
  headers: { 'Authorization': 'Bearer sk_...' },
  body: JSON.stringify({
    userId: callerPhone,
    messages: [{ role: 'user', 
      content: 'incoming call' }]
  })
}).then(r => r.json())
// → "Sarah Johnson, last called Jan 15 
//    about billing dispute, refund processed"

Compatibility

Works with your existing stack

Orthanc is platform-agnostic. It works with any voice AI platform that exposes call transcripts or webhooks.

Vapi

Add persistent memory to Vapi voice agents via function calling or post-call webhooks.

Retell AI

Integrate with Retell's webhook system to store and retrieve memories per caller.

Bland AI

Give cold calling agents memory of previous conversations and prospect details.

Twilio Flex

Enrich call center AI with persistent customer context from every interaction.

ElevenLabs

Add memory to ElevenLabs conversational AI for personalized voice experiences.

Custom stacks

Any voice system that produces transcripts. WebSocket, REST, or webhook integration.

Give your voice AI a memory

Free tier includes 5,000 credits. No credit card required. Integrate in under 5 minutes.