Made in Jordan Voice & chat agents
Voice and chat AI agents, Jordanian first
Muwakil AI is a platform for support and sales agents that understand Jordanian Arabic and reply by voice or chat. Pricing is in JOD, dialects include Jordanian, Saudi, and Lebanese, and recordings stay on our servers or yours.
The studio and the live-call demo are available in the preview environment. Public billing will be enabled at launch.
Muwakil AI, in one frame
Speech in both directions, a dialect voice library, and a live call preview. Pick a tab and press play — no sign-up required.
Live engine is temporarily in sample mode while training runs — live synthesis & recognition turn on automatically when ready.
Your Arabic transcript will appear here after recording or playing the sample.
One platform, every agent tool
Voice, chat, a no-code studio, and an API for teams that already have their own stack.
Voice agents
A real-time call path measured at ~0.37 s median reply latency, with instant barge-in when the caller speaks.
Self-hosting is the default
Models and recordings stay on our servers or the customer’s. This is not an enterprise-only upsell.
Chat agents
Text replies grounded in your knowledge base, with tools, memory, and a handoff to a teammate when needed.
Standard /v1 REST API
The familiar /v1/chat/completions and /v1/audio shape, with a separate key per product.
curl -X POST /v1/chat/completions \
-H "Authorization: Bearer $KEY" Build your agent visually — from first ring to last reply
An Arabic-first board to set voice, dialect, knowledge, and the conversation path, then test before you publish.
- Nine node types: start, conversation, tool, condition, transfer, end and more.
- Instant test runs: the path the agent took lights up on the canvas.
- Safe publishing: publish is blocked while the flow has errors.
From the first word… to the final summary
Listens
End-of-speech detection and dialect-tuned Arabic speech recognition.
Understands
A self-hosted dialogue engine with a private knowledge base per agent.
Acts
Tools and webhooks such as order lookup, with a hold phrase only during tools.
Speaks
One consistent Arabic dialect, streamed before the sentence finishes.
Hands off
Transfer to a human with a summary and a full transcript.
Sits inside your stack, deploys on your terms
Choose how your voice AI runs
Security & controls →Launch in minutes with zero server maintenance, isolated API keys, and per-tenant knowledge bases.
Encrypted transport · JOD plans · Post-call sentiment & summaries included.
Run the entire stack inside your private cloud tenancy connected directly to your internal APIs.
Private peering · Tenant IAM · Direct internal database access.
Full GPU deployment inside your own data center—audio, transcripts, and logs never leave your building.
100% data residency · Zero external AI dependencies · Built for regulated enterprises.
Simple pricing, in JOD
Annual billing, shown per month. Starter, Enterprise and full details are on the pricing page.
- 85 voice minutes
- 500 conversations
- 2 concurrent calls
- 850 voice minutes
- 5,000 conversations
- 5 concurrent calls
- 3,500 voice minutes
- 20,000 conversations
- 10 concurrent calls
Three products… one family
Muwakil AI is the agent engine, alongside the Muwakil Caller contact center and the MuwakilDesk help desk.
Dialect-native voice & chat agents with a visual canvas and API.
Open the studio →An Arabic smart contact center: IVR, queues and reports.
muwakilcaller.com ↗An omnichannel help desk: tickets, social inbox and analytics.
muwakildesk.com ↗Common questions
How fast does the agent reply on a voice call?
Measured median call reply latency in our benchmark runs is ~0.37 s with clause-level audio streaming (~70 ms first chunk).
Which Arabic dialects and voices are supported?
Jordanian, Saudi, and Lebanese—with 8 preset voices plus consented custom voice cloning from a 3–6 second recording.
Can I test Text to Speech and Speech to Text before signing up?
Yes. Use the interactive studio right on our homepage to test TTS, STT, and preset dialect voices with no account required.
How does sentiment analysis work and does it cost extra?
Turn-by-turn mood tracking detects frustration and escalates when needed, while post-call summaries and JSON extraction are included inside the same conversation.
Can I connect my CRM or bring my own API key?
Yes. Connect REST webhooks, MCP servers, and documents in the studio—or run with your own provider key at zero AI markup.
How do teams build multi-step call workflows?
Use the 9-node visual Agent Studio (dialogue, tools, routers, KB lookup, human handoff) with live simulation and path highlighting.
Where does call audio and transcript data live?
In our isolated managed cloud, inside your Private VPC, or fully self-hosted on-premise on Enterprise so audio never leaves your infrastructure.
Where it fits
Stores, clinics, restaurants, real estate, and support — with transparent JOD packages.