Fair Use Policy

Octri runs AI generation, SDK builds, hosted documentation and monitoring ingestion on shared infrastructure. Each of those costs real money per unit of work, so every plan carries allowances. This page says what they are, how they behave at the edges, and what we do when an account goes past them.

The short version: we would rather throttle a feature and email you than take your documentation offline. Suspension is a last resort and never the first thing you hear from us.

This policy forms part of the Terms of Service.

These are the limits in force today. They are read from the same configuration the platform enforces, so this table cannot drift from the product.

Endpoints are counted across the specs in a project. SDK languages beyond your plan’s bundle are available as add-ons at the price shown on the pricing page. Enterprise limits are set in your order form.

Credits meter the AI features. Generating documentation for one endpoint costs 1 credit, and one Ask-the-docs answer costs 1 credit. Nothing else consumes them: editing docs by hand, building SDKs, hosting your site and receiving monitoring data are all outside the credit system.

On paid plans the allowance is restored at the start of each billing cycle and does not carry over. On Free, the 100 credits are a one-time grant so you can try the AI features, not a monthly refill.

Some work is deliberately not charged twice. Regenerating a page we have already generated, where neither your spec nor the content has changed, is served from cache and costs nothing. A failed generation is not billed.

Running out

When the balance hits zero, AI generation and Ask-the-docs stop returning answers. Your documentation stays online, your SDKs keep building, and monitoring keeps ingesting. We warn you by email before you get there.

To carry on, upgrade, or buy a credit pack: 1,000 for $10, 5,000 for $45, 20,000 for $160. Purchased credits never expire and are only drawn down after your plan allowance is used up. You can also turn on auto-refill and let it top you up on a schedule or at a threshold you choose.

Monitoring is measured by the volume of data your application sends us, and it resets with your monthly usage cycle.

We do not cut ingestion off at your allowance. The moment you most need error and trace data is the middle of an incident, which is exactly when volume spikes. Beyond the included allowance we keep accepting data up to 10× your allowance, and track the excess as overage. Only at that ceiling do we refuse a reservation, and we do it with a 429 and a Retry-After header so your SDK backs off and retries rather than losing the batch outright.

Overage above the included allowance may be billed at a published per-gigabyte rate. We will tell you the rate and give you notice before any overage charge is applied to your account. Until then, overage is recorded but not charged.

How long we keep it

Raw events are kept for the window your plan includes, shown in the table above, and are then deleted. Hourly rollups outlive them, so the long-range charts keep working after the individual events behind them are gone. Enterprise windows are set in your order form.

Retention is enforced by the storage engine itself rather than by a cleanup job, so it happens whether or not anyone remembers to run anything. You can ask us to erase an environment sooner; see the DPA.

If you are consistently over, the cheaper answer is usually a higher plan or a sampling rate on your instrumentation. Ask us and we will look at your usage with you.

Every API surface is rate limited. Limits are enforced per fixed window, and a rejected request returns HTTP 429 with a Retry-After header telling you how many seconds to wait. Handle it by backing off; our generated SDKs already do.

• Dashboard and API requests are limited per client. The default ceiling is 100 requests per minute.
• API-key traffic is limited per organisation, and the ceiling scales with your plan. Enterprise is set in your order form.
• Public documentation, changelog and MCP endpoints are limited to 300 requests per minute, which is well above what a browsing reader generates.
• Ask-the-docs is limited to 20 messages per hour per reader, per project, by default. You can raise or lower it per project.

These are guardrails against runaway loops and scraping, not a quota you are meant to budget against. If a legitimate workload needs more headroom, write to support@octri.dev and tell us the shape of the traffic.

SDK generation is queued work: ten languages compiled, verified and packaged per build. We do not cap the number of builds, and we ask for something in return.

• Build when your spec changes, not on a timer. A cron job rebuilding an unchanged spec every few minutes is the pattern we will contact you about.
• Use the GitHub sync or webhooks to trigger on real change events rather than polling.
• Publishing runs on your registry credentials. Registries have their own rate limits and their own rules about republishing versions. Breaking theirs is your problem to fix with them, and repeated failures may pause auto-publish on our side.

Under heavy load we may queue your builds behind others. Enterprise capacity commitments are agreed separately.

Hosted documentation has no published bandwidth cap, and normal traffic to a documentation site, including a launch day or a front-page spike, is exactly what the service is for.

What is not covered is using the docs host as general-purpose file hosting or a CDN for assets unrelated to your API documentation, or serving traffic that is automated at a scale no readership would produce. If your site draws sustained bandwidth far outside what documentation generates, we will get in touch and work out a plan that fits rather than silently throttling you.

Plan limits are per organisation, and the Free and Starter plans are deliberately sized for a single project. Spinning up several free organisations to give one codebase several projects’ worth of allowance is circumvention, and it is the one thing on this page we actively look for.

The patterns that get an account reviewed:

• the same repository connected from more than one organisation;
• one person owning several free organisations that work on the same codebase;
• accounts whose email addresses resolve to the same mailbox, for example the same Gmail address with dots or plus-tags rearranged;
• signups on disposable email domains;
• free organisations whose real AI cost has run far past what the free allowance is worth.

Review means a person looks at the account. No feature is ever switched off automatically on the strength of these signals. Plenty of legitimate accounts trip them: agencies and contractors genuinely run many organisations, and a shared office IP means nothing on its own. Tell us your situation and we will mark the account so it stops surfacing.

Adding real teammates to your organisation is not circumvention. Seats are not what we meter.

We work through these in order, and stop as soon as the problem is resolved:

1. You see it first. Usage is shown in the dashboard, and we email before an allowance runs out rather than after.
2. The specific feature stops. Out of credits stops AI generation and chat. Over the endpoint limit blocks new endpoints. Everything else keeps running.
3. Throttling. Sustained traffic past a rate limit gets 429 responses with Retry-After, never a silent drop.
4. We contact you. For anything that looks like circumvention or a workload we did not anticipate, a person emails you before we take action.
5. Suspension. Only for unpaid balances after contact, deliberate circumvention, or usage that threatens the platform. We give notice unless the issue is a live security or legal problem.

If a limit is in your way, or you think we have got something wrong, email support@octri.dev with your organisation name and what you were trying to do. We would rather adjust a limit for a real workload than have you work around it.

For volumes past the published plans, or for a capacity commitment in writing, talk to sales@octri.dev.

We may update this policy as the platform changes. Material reductions to what a paid plan includes come with at least 30 days’ notice to account owners.