Live — OpenAI-compatible, streaming, metered

Every AI capability.
One grid.

AI Grid is the grid where teams call every model through one OpenAI-compatible API — try them side by side in the playground, compose your own products in the sandbox, and see every call attributed and charged.

playground — no API key neededbuild your own products in the sandboxpersonal workspace on first sign-in
One gateway, two jobs

Ship fast. Stay in control.

Developers want one base URL and a key that works. Whoever pays wants a ceiling, an allow-list and a bill they can read. AI Grid is the same gateway answering both, on every single call.

BUILD

One base URL. One key.

Change base_url and keep your SDK, your tools and your streaming. Keys are aig_…, bound to exactly one project — and that project is the charging context of every call the key makes.

  • OpenAI-compatible chat completions, with streaming
  • GET /v1/models returns exactly what your key can call
  • Playground and sandbox studio, built into the console
See the playground
GOVERN

Assign, cap, account.

An Owner or Administrator assigns a published model version to one exact project. Nothing cascades — team membership alone grants nothing, so a project grant is explicit or the call is refused.

  • Per-project wallets you fund on purpose
  • Allow-lists and daily or monthly ceilings, enforced per call
  • A refused call is never charged
See the controls
Playground & sandbox

You don't just call models. You try them, then you build with them.

Two surfaces sit inside the console, and neither one asks you to wire anything up first. The playground runs on your signed-in session — no API key, no setup. The sandbox studio turns the models a project already has into your own product, deployed and callable through the same API you were going to use anyway.

DRAFT REVISION DEPLOYED CALLABLE
  • Playground — four models, one question. Ask once and compare up to four side by side, each with its own answer, charge and latency. It runs on your session, so there is no key to mint and nothing to install — pick the project that pays and go.
  • Sandbox — compose from what you already have. A node graph over the models assigned to the project, with bounded operators, typed ports, saved revisions and reference file uploads. Deploy one revision and it goes live for that project.
  • Then call it like any other model. A deployed sandbox product shows up in that key's GET /v1/models and answers /v1/chat/completions when its contract is one text in and one text out — anything else runs on /v1/sandbox/{id}/invoke, or as an MCP tool. It stays private to its project throughout.
Open the console
The parts nobody demos

Ship the model. Own the meter.

Every model your teams call, from one host on one bill — with the allow-lists, ceilings and per-call records to run it across an organization.

Every model behind one integration.

Frontier and open-weight models behind one OpenAI-compatible API, so switching model means switching a string, not a client. One stable host, forever: api.aigridapp.com.

client = OpenAI(
base_url="https://api.aigridapp.com/v1", # ← the only change
api_key=os.environ["AIGRID_API_KEY"],
)
client.chat.completions.create(model="openai/gpt-oss-20b", ...)
change one line — keep your SDK, tools and streaming

Money lives on the project.

Every project has its own wallet, funded by the organization Owner. There is no org balance to allocate and nothing to move between levels — you bound a workload by funding its project.

checkout-copilot · wallet41.20 of 50.00
held by open reservations1.80
settled today6.98

Your models, your rules.

Name the models a project may call and everything else is off — an allow-list, not a blocklist you have to keep up with. Add a daily or monthly ceiling, a max output size and a content policy on top. Enforced on every call, not in a wiki.

openai/gpt-oss-20bON THE LIST
everything elseREFUSED
checkout-copilot · 5.00/dayCEILING

Organization, team, project.

Teams own projects; projects hold the wallet, the keys, the assignments and the usage. Membership in a team grants nothing on its own — someone has to grant you the project.

meridian · organization ├─ Team · Platform │ └─ Project checkout-copilot · 41.20 available └─ Team · Data └─ Project nightly-enrichment · 12.05 available

Keys that can't wander.

A key is bound to exactly one project, shown once, and expires in 90 days by default. Give it a model list and a spend cap with a daily, weekly, monthly or lifetime reset. Rotate, and the old key is revoked in the same transaction that mints the new one — no lingering credential, ever.

aig_2b8f····revoked on rotation
aig_7e41····checkout-copilot · 90d · 25.00/mo

Metering down to the token.

Every call lands as a record: time, project, key, product, modality, status, tokens, latency and charge. Reserved at request time, trued-up when the provider answers, released if it cost less — and a call refused before it reaches a provider costs nothing at all.

peak · checkout-copilot · 88k tokens MonThuSun
From zero to first call

Three steps. One afternoon.

No procurement maze, no sales call. Sign in, fund a project, point your SDK at the grid.

01

Sign in — the workspace is already there

There is no registration form. Your first sign-in provisions a personal organization with you as Owner, a team, a project and a wallet. Joining a company organization is an invitation link from its Owner.

app.aigridapp.com
02

Fund the project

Wallets sit on projects, and a new project starts empty — calls answer with an error until the organization Owner tops it up.

402 insufficient_budget
03

Point your SDK at the grid

Swap one base URL, use a key bound to the project that pays, and ship. GET /v1/models tells the key exactly what it may call.

api.aigridapp.com/v1
1base URL — api.aigridapp.com/v1, forever
4models side by side in the playground — no API key needed
100%of spend attributed — project, product, key and call
0charged for a call refused before a provider sees it

Plug in.

Sign in, fund a project, and make your first call before the coffee goes cold.

personal workspace on first sign-in · no procurement · leave whenever