Company
The memory layer
for voice AI
We're building the infrastructure that makes voice AI agents remember. Every call, every caller, every detail — persistent memory for the entire voice AI ecosystem.
Mission
Make voice AI that actually knows its callers
Voice AI is exploding. Vapi, Retell, Bland, ElevenLabs — these platforms make it easy to build AI agents that talk. But none of them solve the memory problem: every call starts from zero.
When a patient calls back, the AI doesn't remember their last appointment. When a sales prospect returns, the AI re-introduces itself. When a customer says "I already explained this," the AI has nothing to reference.
We're fixing this. Orthanc provides persistent, evolving memory for any voice AI platform. Two API calls, sub-200ms latency, and your voice agents remember everything.
Why Now
The voice AI market is ready for memory
$12B+
Voice AI market by 2028
The conversational AI market is growing 24% annually. Every player needs persistent memory, and none have it built in.
0
Voice platforms with built-in memory
Vapi, Retell, Bland — none offer persistent cross-call memory. Their customers are asking for it. We provide it.
200ms
The voice latency barrier
Memory retrieval must be under 200ms or conversations break. We're the only memory API built specifically for this constraint.
Principles
How we build
Voice-first, always
Every feature is designed for the unique constraints of voice: sub-200ms latency, messy transcripts, multi-call journeys, and per-caller identity.
Two API calls, not twenty
Sync memories after calls. Retrieve before calls. That's the entire API. We handle extraction, deduplication, search, and contradiction detection.
Privacy by design
Per-caller data isolation. No cross-contamination. No training on your data. Delete-on-demand. Built for industries that handle sensitive conversations.
Platform agnostic
We work with Vapi, Retell, Bland, Twilio, and custom stacks. No vendor lock-in. Switch platforms and keep your memories.