Concept
How OpenTokenMarket works
The full path from a sponsor top-up to a builder's API call, and how spend flows back.
OpenTokenMarket connects two sides: sponsors who fund AI inference credits and builders who spend them on open projects. A shared gateway sits in the middle, meters every request, and reports spend back to both sides. This page traces the whole path.
Who's who
| Role | What it is |
|---|---|
| Builder | Maintainer of a listed project. Claims a budget, gets one API key, ships against frontier models for free. |
| Sponsor | Person or company that tops up credits and allocates them to projects, with per-project caps. |
| Gateway | A shared LiteLLM proxy in front of every provider (OpenAI, Anthropic, Gemini, DeepSeek…). One endpoint, one key format. |
| Allocation | A slice of a sponsor's balance attached to a specific project, claimable by its maintainer for one week before it expires back to the sponsor. |
The path, end to end
- A sponsor adds credits. A one-off top-up or a subscription converts money into spendable AI budget on the sponsor's balance.
- A project and a sponsor connect — one of two ways.
- Sponsor-initiated: the sponsor finds the project on Explore and adds funds to it directly.
- Builder-initiated: the builder sends a token request to a sponsor, who approves it by hand or through their AI auto-evaluator.
- The budget is allocated to the project with a per-project cap. The allocation is claimable for one week — if the builder doesn't claim it within 7 days, the budget returns to the sponsor.
- The builder claims it. On the project control panel
(
Dashboard → My Projects → your project), the maintainer copies the Project Access Key and base URL from the Project Access Key card. - The builder points a tool at the gateway. Two values — base URL and key — wire up Claude Code, Cursor, Codex, Gemini CLI, or any OpenAI-compatible client.
- Requests run against the sponsor's balance. The gateway resolves the model name to an upstream provider (see the model catalog), forwards the call, and meters the cost.
- Spend syncs back. Usage flows into the app: the builder sees day-by-day and per-model spend on their control panel; the sponsor sees budget consumed against milestones shipped.
- Budget runs out →
429. When the funded balance for a project is spent, the gateway returns429/budget has been exceeded. The key stays valid; the sponsor tops up and requests resume.
What each side controls
- Sponsors control which projects get funded, how much (per-project caps), and whether approval is manual or automatic. Unclaimed allocations always expire back to the sponsor after one week.
- Builders control which models they call and which tools they use. They never see a provider bill and never rotate keys when budget is topped up.
- The gateway enforces the caps and rate limits, keeps model names stable even if the upstream provider changes, and holds the only real provider credentials.
Common questions
Who pays for the tokens?
The sponsor. Builders never pay for inference — the funded allocation covers it.
Does the builder need provider accounts?
No. The gateway holds the upstream accounts. The project access key is the only credential a builder needs.
How does a sponsor know their money did something?
Every project page shows spend against milestones, and the sponsor dashboard aggregates it across all backed projects.
What keeps one project from spending everything?
The per-project cap set at allocation time. The gateway stops that project at 429 when
its share is gone, independent of the sponsor's remaining balance.
How long does a builder have to claim an allocation?
One week. If a project doesn't claim a sponsor's allocation within 7 days, the budget returns to that sponsor automatically.