About
One place to build with AI models.
Relaykit is a small, independent developer platform. It gives you one API key for 11 models from Claude, GPT, Grok and DeepSeek, tested setups for the coding agents you already use, free tools that run in your browser, and writing that keeps you current.
What it includes
- 01GatewayOne OpenAI- and Anthropic-compatible endpoint with prepaid credits and per-key limits.
- 02ChatTalk to every listed model in the browser, billed from the same balance as your keys.
- 03Agent KitsTested setups for Claude Code, Codex CLI, Cursor, Cline and other coding agents.
- 04Free ToolsCost calculator, agent config generator and token estimator. Run in your browser.
- 05ResourcesGuides, tutorials, AI news with sources, and dated product updates.
Why it exists
Building with AI tools meant a separate account, key and dashboard for every provider, and translating between two incompatible request formats whenever a tool only spoke one of them. Working out what a single agent session cost meant exporting spreadsheets from several billing pages.
We wanted one key to point every tool at, one balance to watch, and a receipt for every request. So the gateway speaks both the OpenAI and Anthropic formats, and meters each call at a rate you can read on the models page.
The rest grew from the questions people asked next. How do I set up Claude Code or Codex CLI with a custom endpoint? That became the agent kits, covering 8 tools. What will this workload cost? That became the free tools. What changed this week? That became Resources.
| Input · 4.4k × $2.20/M | $0.00968 |
|---|---|
| Cached · 8k × $0.22/M | $0.00176 |
| Output · 950 × $11/M | $0.01045 |
| Deducted | $0.02189 |
Illustrative token counts at today’s published rate for Claude Sonnet 5. Your usage log shows this breakdown for every real request.
How we work
Five rules we hold ourselves to
Each one is something you can check, not a slogan.
- 01
Prices you can check
Every rate is published per million tokens on the models page. Every request lands in your usage log with its token counts and the exact amount deducted.
Models and rates - 02
Prepaid, so no surprise bills
You buy credit first and requests draw it down. When the balance reaches zero, requests stop. Each key can also have its own spending cap.
How pricing works - 03
The model you name is the model that runs
Requests go to the model ID you send, and that ID is recorded next to the cost. We do not swap in a cheaper model behind the same name.
API reference - 04
Changes are dated and public
New models, rate changes and removals are written up as product updates with a date, so you can see what changed and when.
Product updates - 05
Support is a person
Email support@relaykit.dev and someone who works on the gateway reads it. We aim to reply within one business day.
Contact
Fine print, up front
What we don’t do
Trust is easier to keep than to win back. These are the limits we set for ourselves, and the terms say the same.
- Not a model provider
- We are not affiliated with or endorsed by Anthropic, OpenAI, xAI, DeepSeek or any other provider. We buy API access and pass it through. Model names are their owners’ trademarks.
- No training on your data
- We do not use your prompts or outputs to train models, and we do not sell them. By default we store request metadata only: model, tokens, cost and time.
- No subscription required
- Pay as you go with one-time credit packs: nothing renews, no seat fee, no monthly minimum. Optional plans are opening soon.
- No invented proof
- No customer counts, logos or testimonials we cannot back up. If a number is on this site, it comes from our rate card or settings.
- No uptime promise
- We do not publish an uptime percentage. The status page runs a live check from your browser instead.
- No ads where you work
- Ads, if enabled, appear only on Resources articles. Never in the console, the docs or API responses.
Try it on your own workload
Top up $3 or more, point a tool at the gateway, send a request, and read the receipt.