Argand
Building · 2026
A scheduler for agent work

Every AI turn has to run somewhere. Argand decides where.

Five things get chosen every time an agent does anything — the folder, the hardware, the model, the harness, the cost. Today you pick them by hand, or you don't get to pick at all and everything lands on the most expensive option.

Not an IDE. Nobody sits in it.
It sits underneath the tools you already use.

Re Im 1 i
The five axes

What gets decided, every turn

01

Folder

Which repository or workspace the turn actually operates in.

repo · worktree · scope
02

Hardware

The machine underneath. Selectable, not something you degrade into.

laptop · server
rented GPU · none
03

Model

Frontier when the work needs it. Open weights when it doesn't.

frontier API
local open weights
04

Harness

Which agent runtime wraps the turn. The axis nobody else automates.

Claude Code · Codex
Gemini · raw API
05

Cost

Which tier the spend lands on — with a ceiling that holds.

metered · capped
Why it matters

Most agent work is cheap work

Extraction, drafting, classification, routing, screening. None of it needs a frontier model — but almost all of it gets one, because nothing is deciding otherwise.

Everything hits frontier Underwater

At any plausible seat price, routing every turn to the most capable model available costs more than the seat. This is the wall every agent platform is currently walking into.

Cheap turns, cheap silicon ~95% / ~14%

Stanford's FrugalGPT work reports roughly 95% of frontier quality at about 14% of the frontier calls. The mechanism is published. As of today, no funded product does it automatically.

The gap

The market split in two, and neither side crossed

Model routers

  • Route across model and provider APIs
  • Deep catalogues, real traction, well funded
  • Will happily call a local endpoint — as one more URL
Blind to the machine underneath

Compute schedulers

  • Place workloads across clouds and spot markets
  • Gang scheduling, failover, GPU monitoring
  • Excellent at deciding where code runs
Blind to model and to task
Nobody arbitrates both

And nobody automates harness selection at all — the closest tools make it a setting you configure once, not a decision made per turn. Argand is the join: one control plane, five axes, resolved together.

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. 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 routing layer runs in production today — on one operator's system, which is the point of saying so.

Running now
  • Per-turn execution target across folder, 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
Not yet
  • Local model serving and rented-GPU provisioning
  • The spend ceiling
  • Machine-enforced authority per turn
  • Automatic selection — the part that matters most
Early access

It’s being built in the open.

No waitlist theatre and no launch date. If you run agent work at enough volume that the bill has started to matter, that’s the conversation worth having now — while the routing policy is still being shaped by real workloads.

 

Alpha. Invite only. Nothing has been validated on anyone but the operator — and that’s the first thing we’re trying to fix.