Hakim has been quietly in production for a while.
Today we're flipping the switch on the public side of that: a self-serve tryhakim.ai, an open signup flow, a free tier, an SDK for Node and Python, and a studio you can poke at without talking to a salesperson. This is the moment we come out of stealth.
It's also the part of the arc where a founder letter usually leans on adjectives. "Ambitious." "Category-defining." "Foundational." We'd rather just tell you what we've been doing.
Eighteen months, head-down
We incorporated in May 2024 with a four-person team, two researchers, two platform engineers, and exactly one thesis:
Arabic-first voice AI is a category, not a locale. If you build it that way from day one, it ships.
We didn't raise a mega-round. We didn't announce ourselves. We ran closed alphas, recorded native-speaker listening sessions, and shipped model versions monthly to a small cohort of design-partner teams. The first internal synthesisHakim Fast v0.9, ran in November 2024. The first production APIHakim Fast v1.0, in private alpha, shipped February 2025. The first STT modelHakim Arab v1, came in April.
Through 2025 we iterated: v1.1 was a prosody pass. v1.2 shipped streaming TTS and our first SDKs. v1.3 got us to the 113 ms end-to-end p95 we publish today.
We kept it private because we wanted to be sure. Arabic-first voice is hard to get right, and we didn't want to ship a thesis with a model that didn't support it.
What being "out of stealth" means
A few concrete things change today.
Self-serve is open. You can sign up at tryhakim.ai, get a key, and make your first call without sending an email to anyone. The free tier gives every workspace enough monthly minutes to ship a real prototype before committing to a plan.
Our design partners are named. The first cohort, six enterprises across MENA and Europe, moves from "private alpha under NDA" to "customers we're proud to talk about." They've been patient with our monthly drops for a year; they're the reason the v2 STT model exists.
Regions live today: Frankfurt and UAE. Every call is served from the region closest to the user, your user in Dubai doesn't route through Northern Virginia for a 100 ms synthesis. KSA and Qatar follow in the first quarter of 2026.
The roadmap is public. The /changelog lists what we've shipped going back to founding. The models page publishes our dialect coverage, WER numbers, and the eval methodology. If we regress a metric, it'll show up there before a customer finds it.
What stays the same
The thesis. The team. The pace.
The same four founders are still on the stack. The acoustic decoder in Hakim Fast v1.3 is a continuous evolution of the one we built in the first nine weeks. The dialect coverage in Hakim Arab v2 builds on corpora we started collecting in 2024, we don't ship voices we haven't had native speakers listen to.
Being public is a milestone. It's not a restart.
What's next
In the near term: Hakim Arab v1.5 beta is opening up to the public API (it's already live for design partners). KSA and Qatar regions light up in Q1. Voice cloning, currently consent-gated beta, moves to GA when the audit trail tools are done.
Longer term: more dialects (Moslawi, Omani, Libyan are in research), more languages in the MENA tier, more real-time features, and the kind of low-level platform work (quotas, observability, SDKs, CLIs) that separates a model from a product.
If you've been waiting for a reason to try us, this is it. The signup takes thirty seconds. Tell us what you ship with it.
— The Hakim team