zynlab
A customer asks, anywhere
A message arrives from your web widget, a QR code, or the phone. Session scope — which account, which location, which customer — rides along automatically.
Your business, answered.
Zynlab is the multi-tenant agent backend any business plugs into. It grounds every reply in your knowledge, calls your systems to actually get things done, and talks to your customers on the web, in a widget, and on the phone.
Drop-in compatible with OpenAI /v1/chat/completions and
Anthropic /v1/messages
Z
Concierge
Room 412 · Web & Voice
Hi! We're out of towels — and could we get two extra pillows? 🙏
Of course — fresh towels and two extra pillows are on their way to Room 412.
Anything else I can take care of?
Dispatched
→ housekeeping
towels ×2
pillows ×2
audit #0412
Knowledge-grounded
Calls your APIs
Voice with barge-in
One-script widget
Multi-tenant
Human escalation
Audit trail
From a message to a done thing.
Grounds & decides
The engine retrieves from your tenant's knowledge base and reasons over a curated, tenant-scoped toolset — never inventing what it wasn't given.
Acts on your systems
Declarative HTTP tools call your booking, CRM, ticketing, ATS, or PMS — dispatching a structured action with an idempotency key and an audit trail.
Confirms & escalates
The customer gets one clear sentence back. Complaints, edge cases, and anything unclear are handed to a human — never silently auto-actioned.
One engine. Any industry.
The concierge
Room service, housekeeping, late checkout, amenities, local tips — dispatched to the right team, 24/7.
The front desk
Appointments, hours, departments, directions — and a firm hand-off to clinicians for anything medical.
The hiring assistant
Open roles, requirements, application status, and interview scheduling for every candidate, tirelessly.
Whatever you run
Retail, SaaS support, services, internal help desks. Set the persona, point it at your knowledge, wire your tools.
The same assistant, four different hats — switch between them live in the demo.
Production depth, not a chatbot demo.
Knowledge grounding
Per-tenant RAG over your policies, catalog, hours, and docs. Answers cite what you fed it — and stay silent on what you didn't.
Action tools — your APIs
Declarative HTTP executors turn intent into real work in your systems. Onboard a new capability with config, not a code release.
One-script widget
Drop one script tag into your site for a white-label chat & voice widget — keyed by a browser-safe publishable key and an origin allowlist.
Voice with barge-in
A streaming speech pipeline takes calls in a natural, interruptible conversation — the same persona and guardrails as chat.
Conversation records
Opt-in transcripts and call logs with PII redaction on by default, retention you control, and a Messenger-style thread for your staff.
Multi-tenant by design
You own the integrating customer; they pass per-session scope. One engine core, clean isolation, per-key cost tagging.
Voice
It can answer the phone.
The same engine, fronted by a streaming STT → reason → TTS pipeline. Customers call and get a warm, fluent answer — and they can talk right over it.
- Rings, picks up, and speaks — like a real phone line
- Barge-in: speak the moment you want, it stops and listens
- Checks in on silence and ends dead-air calls gracefully
- Fully open-source pipeline — no per-minute STT/TTS fees
If you can call OpenAI, you can call Zynlab.
POST /v1/chat/completions
200 OK · streamed
curl https://api.zynlab.dev/v1/chat/completions \
-H "Authorization: Bearer zl_live_…" \
-H "Content-Type: application/json" \
-d '{
"model": "zynlab",
"stream": true,
"session": { "account": "acme", "location": "downtown" },
"messages": [
{ "role": "user", "content": "Can I move my appointment to Friday?" }
]
}'
# → streams a reply, calls your tool,
# → and returns a structured action + audit id.Bearer-auth per tenant. Same SSE shape as OpenAI — Open WebUI, LibreChat, Continue, and your existing chat UI work unchanged. Full docs →
Sized to your operation.
Starter
- Knowledge-grounded chat assistant
- Web widget, one script tag
- Operator console & metrics
Pro
- Everything in Starter
- Action tools wired to your systems
- Conversation records & escalations
- Realtime ops dashboard
Enterprise
- Everything in Pro
- Voice calls with barge-in
- Live web search
- Integrator multi-tenant provisioning
Per-contract entitlements on every plan — talk to us about yours.
Get started
Put an agent on every channel.
Bring your knowledge and your systems. We'll wire up the brain that talks to your customers — on chat, in the widget, and on the phone.