Concepts

TensorLoop has five things to track. Read this once and the rest of the docs will make sense.

Account

You sign in with email or Google. An account holds:

• One plan (free or pro).
• Up to N keys (2 on free, 10 on pro).
• Spend history.

Accounts are isolated. Other users can't see your keys, aliases, or spend.

Plan

The plan sets ceilings. It is not a per-call setting — it's an upper bound on what you can configure when minting a key.

See Models.

Key

A key is a bearer token. Every inference request carries one. A key is:

• Scoped to your account. Nobody else can use it.
• Scoped to a model allowlist. Set at mint time. Requesting another model returns 403.
• Bounded by a budget and an RPM. Hitting either returns 429.
• Bounded by an expiry date. Default 90 days, max 365.

You mint and revoke keys in the dashboard. There is no per-key rotation — you mint a new one with a new plaintext value.

Budget

A USD ceiling on a key. Resets every 30 days, counted from the key's mint date (not the calendar month). Once a key spends its budget, every call returns 429 Budget exceeded until the rolling window ticks past.

Rate limit (RPM)

A per-key cap on requests per minute, measured as a rolling 60-second window. Bursts past the cap return 429. Set it lower than the plan max to tighten the blast radius if a key leaks.

The two surfaces

TensorLoop has two HTTP endpoints. They authenticate differently and serve different purposes:

You only need to know about the inference surface to build. The dashboard surface is what tensorloop.tech itself uses.

How a request flows

your app
   │  POST /v1/chat/completions
   │  Authorization: Bearer tl_...
   ▼
LiteLLM proxy (litellm.tensorloop.tech)
   │  validates key, checks model allowlist,
   │  checks budget + RPM, meters cost
   ▼
upstream model provider
   │  returns completion
   ▼
LiteLLM logs spend → Supabase → dashboard

Spend lands in your dashboard within seconds. See Dashboard walkthrough for what each panel shows.

What's next

• Getting started — sign up and make a request.
• Authentication — key lifecycle in detail.
• Rate limits & budgets — what each cap enforces.