Limits & billing

Cobble has one billing model: your plan is your flat rate, your wallet covers everything past it. No meters inside your plan, no surprise overages outside it. This page explains exactly how that works — with real numbers, because vague fair-use policies help nobody.

The plan: unmetered requests inside a rolling window

Each plan includes a number of chat requests per rolling 5-hour window:

Inside the window, requests to your plan's models are unmetered — we don't count your tokens, and a long conversation costs the same as a short one. The window rolls continuously: a request made at 9:00 stops counting against you at 14:00. There are no daily cliffs and no monthly buckets to ration.

What counts as one request: one call to /v1/chat/completions. Agent tool-call loops make one request per round, so a long agentic coding session can use 100+ requests — worth knowing when picking a plan.

Slots: how many requests can run at once

A slot is one request executing concurrently. One slot means requests queue behind each other; two slots mean your agent's parallel tool calls actually run in parallel. Agent Runner's slot is pinned — reserved capacity, so a long-running agent gets consistent latency instead of re-entering the queue on every call.

Extra slots are available as add-ons on the Coding plan.

The wallet: everything past the flat rate

Your wallet holds dollar-denominated credit from top-ups ($10–$100, or custom). Top-ups never expire. The wallet pays for three things, all at the published catalog rates:

1. Spillover — requests beyond your window, if you enable it
2. Models above your plan — try the flagship before upgrading
3. OCR and embeddings — always wallet-billed, per page and per token respectively

Spillover: what happens when you hit the window

You choose, in the dashboard:

• Spillover off (default): requests past the window return 429 with a Retry-After header until the window rolls forward. Nothing is ever billed beyond your subscription.
• Spillover on, with a cap you set: requests past the window keep working, billed from your wallet at catalog per-token rates, up to your monthly cap. Your agent never hits a wall mid-session unless you decided where the wall goes.

There is no way to spend wallet money you didn't explicitly top up, and no way to exceed a cap you set. That's the whole policy.

Checking your usage

The dashboard shows one view: your current window usage, your slots in use, and your wallet balance. Programmatically, rate-limit state is returned on every response:

X-Cobble-Window-Limit: 750
X-Cobble-Window-Remaining: 512
X-Cobble-Window-Reset: <seconds until oldest request rolls off>

Plan changes

Upgrades take effect immediately; downgrades at the next billing cycle. Lapsed subscriptions don't kill your keys — they fall back to wallet billing at catalog rates, so nothing you built stops working.

Fair use, in numbers

We don't have a hidden fair-use policy — the window and slot limits above are the policy, enforced mechanically. Things we do prohibit: reselling raw plan capacity, sharing one subscription's key across an organization (that's what team seats are for), and circumventing per-account limits with multiple accounts.