Why we are Wakir Labs
Most companies pick a name and move on. We spent five rounds and around fifty candidates on ours, and this essay records why — not because the name is unusually clever, but because the choice documents a decision we plan to keep making in public.
What “Wakir” actually means
Wakir comes from Arabic wakīl (وكيل). In classical and modern legal Arabic it denotes an appointed representative — a proxy, a deputy, an agent in the legal sense. Not the AI-pop sense (“an agent is a thing that does work for you”), but the older one: someone is named, someone is accountable, and the relationship is documented.
That is the sense we wanted on the door. Most multi-agent frameworks treat humans as an emergency exit — a fallback when the agents get confused. We design the other way around. People are first-class roles in the workflow, with their own primitives for approval, hand-off, and audit. Wakil is the right word for what we build because it forces the question we keep asking: who is this agent acting on behalf of, and how do we know?
We did not pick the word because Arabic was exotic. We picked it because it was precise. Latin, Greek, and Sanskrit roots were on the list too, and most of them collapsed under the weight of existing software brands — Latin and Greek are where the Eclipse and Apache pools live; the USPTO is full of them. Arabic gave us a clean semantic hit on “agent” and a clean live-verification result. Both mattered.
Why not “AI Inc.” or “X Labs”
The names we deliberately avoided are more revealing than the one we picked.
We did not want AI in the name. There are roughly forty companies named “Something AI” launched every quarter; the suffix tells the listener nothing except that the company would like to ride a wave. By the time anyone looks at us seriously, the wave will be a different shape, and the name will read like a fossil.
We did not want a generic Labs without a distinct stem. Mistral Labs, Mantra Labs, Manifold Labs — perfectly good companies, indistinguishable in a paragraph. Wakir is distinctive enough to carry Labs without the suffix becoming the whole brand.
We did not want a Latin compound (Trovus, Vortus, Provex, Trovex). We tried; they all bumped into existing software companies in adjacent fields. There is no point in picking a name your future GitHub-org clashes with on day one. We verified that each candidate failed live, with a real search, before moving on.
We did not want a metaphor we would have to explain in every conversation. “Wakir” needs one sentence of etymology and an audio clip. After that, it is a name, not an argument.
What “accountable multi-agent systems” means in practice
The tagline on our cover is Accountable multi-agent systems, a framework for AI-first organizations. That is a thing we have to cash, not just print. Three concrete implications follow:
Identity-first agents. Every agent in our framework has an identity that is more than a string. It is signed, verifiable, and tied to a defined set of capabilities. You can ask, of any action in the system, which agent did this, what was it allowed to do, and how do I prove that? — and get an answer that does not require trusting our database. The cryptographic mechanics get their own posts; the design choice does not.
Humans as a typed role, not an emergency. A workflow that has no defined human role is, in our framework, a bug. Approval, handoff, clarification, review are primitives with types and contracts, not Slack workflows we hope someone will follow. If you want a fully autonomous agent loop, you can write one — but you have to opt out of the human role explicitly, which is an audit event in itself.
Audit trail as a first-class kernel primitive, not bolt-on logging. Every action — agent or human — emits an event into a stream that is the system of record. The stream is part of the public surface of the framework. The logging is the architecture, not a layer added at the end. This is what lets us anchor decisions in OpenTimestamps and make the kind of belastbare claims that compliance officers actually accept.
None of this is novel in isolation. Distributed systems people have been making this argument for twenty years. What is novel is treating these properties as the first thing the framework provides, instead of the last thing a customer asks for.
What “Labs” buys us
Labs is doing real work in the name. It signals R&D and continuous iteration, not “we shipped a product, now we sell it.” That is consistent with how we plan to operate: an immutable documentation layer (every architectural decision, every postmortem, every failed experiment recorded with timestamp and signature), a publication cadence of several substantial pieces a week, and a willingness to publish tools we tried and dropped alongside tools we kept.
A company whose method is write everything down and time-stamp it is closer in spirit to a research lab than to a SaaS shop. We did not want to call ourselves a research lab — we are not in the business of papers. But we wanted the suffix that says: there is a research practice in here, and it is on purpose.
How we will use the name
Practical, for the record:
- Canonical written form: Wakir Labs. First mention in any material, written out in full.
- Short form: Wakir. Once context is set — in compounds like the Wakir framework, the Wakir platform — Wakir alone is fine.
- Pronunciation: Wa-keer Labs in English, Wa-kir Labs in German. There is a phonetic note on every cover surface where a first-time listener is likely to land.
-
Domain:
wakirlabs.com. Reserve domains exist; canonical address is the .com. We do not run a parallel presence on a second domain. -
Repos:
github.com/wakir-labs/<component>. No brand repetition (wakir-labs/wakir-frameworkis redundant;wakir-labs/frameworkis enough).
That is the brand surface. The substance is everything we publish, every commit we anchor, every ADR we expose. The name is a promise; the rest of this site is the receipts.
We will sign things — Wakir Labs Editorial on essays, individual names where appropriate, signed commits everywhere. Wakil means “the one named to act.” That is the standard we are holding ourselves to.
Below is what that standard looks like as a concrete artefact: a multi-hour audit-trail run with all four independent OpenTimestamps calendars finalised on Bitcoin. The hashes are the receipts, the block heights are the witnesses, the verify command is the invitation.
— Wakir Labs Editorial