EARLY ACCESS
Caffeine Inference is cost-effective AI with smart routing. It runs on your Caffeine subscription — no separate billing, no separate account. Your existing tools can point at it and immediately get smart, economical model selection.
The config is three lines away. No model picking, no key management, no worrying whether a model is busy or slow — the router handles all of it for you.
> WHAT IS CAFFEINE INFERENCE?
Caffeine Inference is Caffeine's AI inference layer. It combines cost-effectiveness with smart routing: you send a request, the router weighs cost, load, and difficulty, picks a model, and returns the answer.
The key point: it runs on your Caffeine subscription. No separate billing, no separate account — you're already paying for it.
// Caffeine Inference is in early access — pricing and routing evolve as it improves.
> SMART ROUTING — HOW IT WORKS
You never choose a model. The router weighs everything for you:
1. REQUEST
You (or your tool) send a request to Caffeine Inference. The request has the same shape as any standard AI API call.
2. ROUTER WEIGHS
The router weighs cost, load, and difficulty to pick the best model for your request — you never choose yourself.
3. PICK MODEL
The router picks a model and sends the request. When a model is busy or slow, the request routes elsewhere instead of failing.
4. ANSWER
You get your answer — no models, keys, or separate billing to manage.
[NO MODEL PICKING]
> POINT EXISTING TOOLS AT CAFFEINE
Claude Code and anything accepting a custom base URL can point at Caffeine Inference. The request shape is the same — you only change the config.
| Tool | Config | Note |
|---|---|---|
| Claude Code | ANTHROPIC_BASE_URL / ANTHROPIC_AUTH_TOKEN | Set the base URL and auth token and Claude Code sends its requests to Caffeine Inference. |
| Goose | GOOSE_PROVIDER / OPENAI_HOST / OPENAI_API_KEY | Configure the provider, host, and API key so Goose points at Caffeine Inference. |
| Other tools | Base URL + key | Anything accepting a custom base URL can point at Caffeine — the request shape is the same. |
> FOUR-STEP SETUP
1. OVERVIEW
Understand what Caffeine Inference is: cost-effective AI with smart routing, running on your Caffeine subscription.
2. CHOOSE A TOOL
Pick the tool you want to point at Caffeine — Claude Code, Goose, or anything accepting a custom base URL.
3. SET UP
Copy the config in the format that tool expects — settings are given in the format that tool expects.
4. TEST
Send one call and confirm the request reaches Caffeine Inference and returns an answer.
> QUICK START PROMPT
Copy this prompt and paste it into your Caffeine project. It configures Caffeine Inference for your project, pointing a tool that accepts a custom base URL:
Add Caffeine Inference to this project. Point my tool at the Caffeine Inference base URL using the custom base URL + key config, then send one test call and show me the answer.> PRICING
Pricing is simple and transparent, subject to change as routing improves:
| Item | Price |
|---|---|
| Input | $0.2 / M tokens |
| Cached input | $0.05 / M tokens |
| Output | $0.5 / M tokens |
> FAQ
> Can I pin a specific model?
By default the router picks the model for you. If you need to pin a specific model you can — but smart routing usually gives the better cost-quality balance.
> What happens when a model is at capacity?
The request routes to another model instead of failing. You still get an answer — just from an available model.
> Does inference spend the same Caffeine budget as apps?
Yes — inference runs on your Caffeine subscription, with no separate billing or account.
> Is it compatible with existing API calls?
Yes — the request shape is the same as standard AI API calls, so tools pointing at Caffeine don't change how they call.
> Are there rate limits?
There are rate limits to ensure fair access for all users. They adjust as routing improves.
> Are there spend caps on a key?
There are spend caps to help you control costs. They align with your Caffeine subscription budget.
> Is my prompt usage used for training?
Your prompts are not used to train models — your data stays private.
// Caffeine Inference Checklist
- I understand what Caffeine Inference is and how it runs on my subscription
- I chose a tool to point at Caffeine
- I copied the config in the format that tool expects
- I sent one test call and received an answer
> GET STARTED
Caffeine Inference runs on your existing Caffeine budget — no separate billing, no separate account. The config is three lines away.
This is early access — pricing and routing evolve as it improves. Start now and bring smart, economical AI into your existing tools.