Spare AI,
put to work.
Nonprofits can now get AI for their staff cheaply. What they still can't get is AI to run a service: thousands of documents, every day, metered and unpredictable.
We fund that gap, deliver the capacity through one simple system, and produce a verified record of exactly what every dollar did. Nonprofits never pay.
A little for everyone,
a fortune for three,
nothing in between.
Every nonprofit can get a small pot of free AI credit. It runs out in weeks if you're doing anything serious. At the other end sit enormous competitive grants that go to novel, well-written, photogenic projects. In between is where almost every organization actually lives, and nothing exists at that size.
A verified nonprofit applies
They describe one specific job: “summarize every case file that comes in.” Not open-ended access. A named piece of work with an end.
We set a budget and a cap
The work runs on our account, through our software. They never hold an API key, never see a credit, and can't overspend the grant.
The record writes itself
Because every request ran through us, we already know what happened. No forms to file, nothing to take on trust.
We start with backlogs, not services, a finite pile of work with an end date. Backlogs finish cleanly, which means no nonprofit is left depending on something it can't afford once the grant runs out.
We never store your documents
Work is processed and discarded. We keep the metadata (how many requests, what they cost, when they ran) and nothing else. The impact record needs “4,038 files, six weeks,” not the files.
You never give us credentials
You sign in to our software. Our software talks to the AI providers on our account. There is no key to hand over, no login to share, and nothing of yours for us to hold.
Nothing is pooled between nonprofits
Your grant is yours. Another organization's usage can't draw it down, and a problem in one organization's work doesn't reach into another's.
You cannot overspend
The limit is checked before each request runs, not reconciled afterwards. When the grant is spent, it stops. No invoice ever arrives, not to you and not as a surprise to us.
These aren't policies we could quietly change. They're consequences of how the system is built, which is the only kind of promise worth making.
It's the first thing anyone asks, and they're right to ask it. There are three ways to run something like this. Two of them will get accounts banned, and they're why most versions of this idea never get built. We do the third.
We're just a customer
We buy AI from the providers like any business, and build something on top of it. That's what every AI company does.
Nothing is shared
No nonprofit gets a login, a key, or a credit. There is no credential to share and no credit to transfer.
The money is a grant
Funders give money, not credits. An ordinary grant, which is well-understood ground for everyone involved.
You have a pile of work and no budget to run it
Case files to summarize, documents to translate, an archive to index, forms to extract. You've read that AI could help and have no way to pay for it at volume. We fund it, and you keep working the way you already do.
You want to fund AI for nonprofits without running a contest
Competitive grants reach a few winners and generate a press release. The same money through us reaches forty organizations, with a metered record of what every dollar did, reported per case processed rather than per unit of infrastructure.
You have AI capacity you're not using
Fund a pool and be named in it. What makes the naming worth something is the record: not a claim that you helped, but a verifiable account of what your contribution actually did.
Marcus Anum
I teach AI to people who've been told it isn't for them. Three cohorts so far with the African Bridge Network in Boston, immigrant professionals rebuilding careers here, plus a City of Boston-funded course this autumn. I write the curriculum, run the labs, and sit with people while it clicks.
Which is how I ended up here. The organizations I teach for could use AI to do far more than help their staff write things. They have backlogs (case files, translations, forms, archives) that would take a person a year and a machine a week. What stops them isn't knowing how. It's that nobody funds the running cost of a service, only the launch of a project.
Sparecycle is my attempt at that specific gap. Before this I was a product manager, and I'm based in Boston.
Course material from the classes I teach is public at github.com/Niimarcus/ai-class.
What's true today
Sparecycle is pre-launch. The gap is documented, the architecture is designed, and the model is legal as built: nothing is pooled, no accounts are shared, no credits change hands. We're a customer of the AI providers like any other business, and nonprofits use our software rather than our credentials.
What we're testing
One question, with nonprofits, right now: is there a job you'd run across everything you have, if the cost weren't in the way? Everything else waits on those answers. If the honest answer turns out to be no, we'd rather find that out in September than after building for a year.
If you run a nonprofit,
we want the boring problem.
Not the exciting pilot. The pile of work nobody has time for. We're talking to organizations now, and those conversations shape what gets built. Funders and companies: same address.
hello@sparecycle.org