BYO compute
Your machineYour data stays on hardware you control. You keep an Argand node running; we operate the control plane.
No hosted compute costArgand
Your work, tools, and AI — one system you can understand
Build an AI system that understands your world. Learn it while you build it.
Argand connects the work you care about to the right models, machines, and tools while keeping consequential judgment yours. Its guided private experience helps you map the system, understand every boundary, and produce a portable recipe for your own Argand.
Build and learn your ArgandNot an IDE. Nobody sits in it.
It sits underneath the tools you already use.
The whole thing, precisely
Argand is an operator‑governed, algorithmically agentic, self‑calibrating, distributed network protocol for carrying state across any runtime and interface.
The half that doesn’t compress. Judgment stays yours.
The half that does. Models, machines, harnesses, price — all racing to the floor.
The protocol has a name: AAP, the Argand Arbitration Protocol. Argand is the product; AAP is what it speaks. A client someone else writes implements AAP — not Argand — which is the whole difference between a layer and an app.
The agent — model plus harness — is selected per task rather than configured. Everything on the real axis gets cheaper every quarter; nothing on the imaginary one does, which is why it’s the half we don’t automate.
That sentence describes where this is going, not where it is. See Where it stands.
The six axes
What gets decided, every turn
Four of them optimise downward. Two optimise the other way — and those two are what make it something a stranger can run.
“Pick the cheapest that works”
01–04
01
rented GPU · API
Runtime
Which machine, whose machine, and which server turns weights into tokens.
laptop · your serverrented GPU · API
02
open weights
Model
Frontier when the work needs it. Open weights when it doesn’t — which is most of the time.
frontier APIopen weights
03
Gemini · raw API
Harness
Which agent loop wraps the turn. The axis nobody else automates.
Claude Code · CodexGemini · raw API
04
State
What the turn resumes from — the workspace, the loaded context, the topic. Carried between machines, not pinned to one.
workspace · context
“Pick the most restrictive that works”
05–06
05
Authority
What this turn is allowed to do. Read, draft, or reach the outside world — enforced by the machine, not by convention.
Not built yet
06
Budget
What it may spend, checked before anything is provisioned rather than after the bill arrives.
Not built yet
How it decides
Bottom rung first
Not “which model is best here” — that’s a prediction problem nobody has solved. Just “is the cheap one good enough,” which is a threshold, and thresholds are answerable.
↑ escalates only on evidencecostlier
01
Your own hardware
Open weights on the machine already sitting there. Nothing leaves it.
$0 marginal
02
Rented silicon
Bigger open models on spot GPU, provisioned for the job and torn down after.
cents / hour
03
Frontier
The work that genuinely needs it — long-horizon reasoning, the hard refactor.
per token
start here, every turncheapest
Starting at the bottom is nearly free to get wrong. A failed attempt on hardware you already own costs seconds, not dollars — so the question isn’t how often the cheap rung is right, it’s how fast it fails when it isn’t.
And nothing gets offloaded that can’t be checked. A test, a schema, a diff, a second pass. The checker is what triggers the escalation — which means the savings are bounded by what’s checkable, not by what’s cheap. We’d rather say that than discover it later.
Why it matters
Most agent work is cheap work
Extraction, drafting, classification, routing, screening. None of it needs the most capable model available, and all of it currently gets one.
Everything hits frontier
Underwater
At any plausible seat price, routing every turn to the most capable model costs more than the seat. This is the wall every agent platform is currently walking into.
The default, today
Routed by task shape
2.3× cheaper
An independent August 2026 study answered more questions correctly than the best single model, at $3.33 against $7.69 — by assigning models per task type rather than picking one. Small study. Same direction as Stanford’s FrugalGPT.
arXiv:2608.23023 · not ours
The gap, honestly
Others have built the parts. Nobody routes.
An earlier version of this page claimed the market had split in two and neither side crossed. That turned out to be wrong, and we’d rather correct it here than defend it.
| Agents | Runtime | State | Routes it | |
|---|---|---|---|---|
| Model routers | ✓ | — | — | — |
| Compute schedulers | — | ✓ | — | — |
| Multi-harness clients | ✓ | partial | ✓ | — |
| Hosted agent platforms | ✓ | one cloud | ✓ | — |
| Argand | ✓ | ✓ | ✓ | ✓ |
The whole difference is the last column
Every product in that table is a picker. You choose the harness, you choose the machine, you choose the model — by hand, every session. Good tools, and not one of them decides for you. That last column is the entire thesis, and it’s the only one we’d defend.
How it’s delivered
Run it yourself, or don’t
Same arbitration layer at every rung. What changes is who owns the machine and who carries the operations.
Open core
Self-host everything. The router, the adapters, the tuple, the AAP spec — plus a setup recipe your own agent runs against whatever hardware you happen to have. Write your own client if ours isn’t the one you want.
free
Bring your own compute
Hosted control plane, your machine registered as a node. Turns on your hardware never leave it.
no compute cost
Shared workspace
On-demand containerised workspace, pooled compute underneath.
seat + metered
Dedicated
Your own box. Isolation, residency, white-label.
premium
Engagement
We build the first workspaces with you, on your own workflows and data.
scoped
Where it stands
Honestly, early
The execution layer runs in production today — on one operator’s system, which is the point of saying so. Everything below is measured or marked unbuilt; nothing is projected.
Running now
- Per-turn execution target across state, tier, harness and model
- Four agent harnesses behind one interface
- Cost accounting by tier
- Records where a turn actually ran, not just where it was sent
- Approvals that fail closed, and admission control
Not yet
- Rented-GPU provisioning
- The spend ceiling — and nothing gets rented until it exists
- Machine-enforced authority per turn
- State that survives a hop between machines
- Automatic selection — the part that matters most
Hosted founding cohort
You choose the workspace. Compute stays a separate dial.
These are workspace options, not GPU bundles. We review every brief and build accepted workspaces by hand. This is not instant.
Founding cohortTaking 5 of 50 provisioned · the counter moves only when a workspace is live
How the offer works
Submitting a brief is free. We review it, reply directly, and build only after agreeing the fit. There is no card hold and no charge for something that does not exist. We invoice only when the workspace is provisioned; the founding rate starts that day and is held for 12 months. Metered compute is separate.
Hosted brief
Tell us what you need to run.
Choose the closest workspace option and describe the work. A person reviews every submission. We do not auto-provision, and this form does not start billing.
Alpha. Five hosted workspaces, built by hand. Submission is free; invoicing begins only on provision.