Usage Quotas

ai.KMITL includes a daily allowance and a weekly platform usage quota for every account. Usage is based on model cost, not message count. The daily allowance is used first; once it runs out, usage comes from the weekly quota.

Included Usage

• Daily allowance, refreshed every day at 09:00 Bangkok time
• Weekly included usage, refreshed every Monday at 00:00 Bangkok time
• All built-in models available while quota remains
• No credit card required
• Bring Your Own Key (BYOK) remains available when you need more usage

How Usage Is Counted

Built-in models deduct what the provider actually billed for each response. That includes a response that fails partway: every step that ran before the error was billed by the provider, so it counts. A request that fails before any model step runs, or that reports no cost, is not charged.

Tools that run on the platform's own models count too: the small model that condenses pages for web search, the reader model that answers questions about long or scanned PDFs (including the automatic transcription of scanned PDFs), and the Deep Research model. They show up in Settings → Usage under those models.

Model selection shows a simple tier:

• $ — Cheap: lowest preflight reserve
• $$ — Medium: moderate preflight reserve
• $$$ — High: highest preflight reserve

The tier sets a before-request reserve and indicates relative usage, not a fixed charge per message. The final deduction is the provider's reported cost. Large attachments, long conversations, and extra reasoning can use more quota even on a cheaper model.

Long Responses Near Your Limit

A response that uses many tools can cost more than the reserve. If a response reaches the quota you had left when it started (or gathers a very large amount of context), the AI stops calling tools and writes its answer from what it already has, so you still get a complete reply. The reply may be a little over your remaining quota; your next request may be blocked until an allowance resets.

Checking Your Quota

The quota indicator shows the combined percentage remaining, not an exact amount. Hover over it to see separate Daily and Weekly bars, each with its percentage remaining and reset time or date. Reset labels use your device's local time.

In Settings → Usage, you can review token activity, request volume, and model breakdowns.

Estimate Before You Send

When a message is likely to use a noticeable share of your weekly quota, an estimate (for example ≈ 4%) appears next to the Send button. It counts the conversation so far, your attachments, and your message, priced from what that model has recently cost. Tool use such as web search or reading files adds to it. For $$ and $$$ models, the estimate offers a lighter model you can switch to in one click.

Reset Schedule

The two allowances reset independently:

Allowance	Reset schedule
Daily	Every day at 09:00 Bangkok time (UTC+7)
Weekly	Every Monday at 00:00 Bangkok time (UTC+7)

Each reset restores that allowance; it does not refill the other one. Unused quota does not roll over.

What Happens at the Limit

When the combined remaining daily and weekly quota cannot cover the selected model's tier reserve:

1. New built-in model requests are blocked for that tier. A $ model may still fit the remaining quota.
2. Existing conversations remain available.
3. You can still manage settings, folders, and previous chats.
4. You can continue with BYOK or wait for the next daily or weekly reset.

Running out of the daily allowance alone does not stop you: the weekly quota is used next. Likewise, a fresh daily allowance can be used even if the weekly quota is exhausted.

Making Your Quota Last

• Start with a $ model. Use it for summaries, translation, search, and everyday questions. If the answer is not good enough, change to a more capable model in the picker, then use Retry on the response. Move to $$$ only for genuinely hard problems a cheaper model could not solve.
• Keep files and chats focused. Every message re-sends the conversation history. PDFs are converted to text and long ones are sent as an outline the AI reads from, but turning on Send full document sends every page again with each follow-up. Keep one topic per chat; switching models in the same chat keeps its history.
• Write one complete prompt. Include your goal, relevant context, and desired format before sending. To correct a text prompt, use Edit instead of stacking correction messages. Edit replaces that prompt and regenerates from there, removing later messages; it still uses quota and does not refund the earlier response. Messages with attachments cannot be edited.
• Use low reasoning effort for simple tasks, where the model picker offers that control. Raise it only when the problem needs deeper reasoning.
• Turn on web search only when you need current information. Leave it off for tasks such as rewriting, translation, or questions about the file you supplied.
• Use Continue for a cut-off answer, when offered, instead of Retry. It resumes the unfinished response rather than repeating the work; the continuation still counts toward quota.

Use a Short Digest with a Premium Model

For a large document where you need synthesis rather than exact wording:

1. In chat A, use a $ model to extract the relevant information into a workspace file such as notes.md. Ask it to preserve key numbers, short quotes, and page numbers, not just write a broad summary.
2. Download the notes file. Workspace files belong to each chat; they do not transfer automatically.
3. Open a new chat on the premium model and attach only the notes, not the original document.

Check the Source

A digest loses detail and may omit important information. Check key claims against the original. This workflow is not suitable for exact-clause lookup; attach the relevant original pages for that task.

BYOK

BYOK requests use your own provider key and do not deduct the included platform quota. See BYOK Guide for setup.

Economy Mode

When the platform is running on a tight budget, admins can turn on Economy Mode to keep the service available for everyone. While it is on:

• A banner at the top of every page says so, and you'll get a one-time notice explaining the change.
• Built-in requests run on the Economy Mode models (for example "kAI Economy"), listed at the top of the model picker. Other built-in models are hidden; Show unavailable lists them greyed out.
• If you send a message, retry, or edit with a model that isn't available, ai.KMITL switches to an Economy Mode model and shows a short notice saying which one.
• Image Studio and Movie Maker are paused.
• Usage still counts toward your quota as usual.

Models you can reach with your own API key keep working normally. See BYOK Guide.

Fair Use

The included quota is for individual academic and work use. Automated bulk processing, account sharing, reselling access, or system abuse are not allowed.

Account Sharing

Each account is for individual use only. Do not share your login credentials.

Strict Consequences

Any user found sharing accounts or violating the fair usage policy may have their account suspended.

FAQs

Can I buy more quota?

Not currently. Use BYOK for additional usage or wait for the next daily or weekly reset.

Do failed responses count?

Only for the work that ran. A response that fails partway is charged for the steps the provider billed before the failure; a request that fails before any model step runs is not charged. If a response stopped before its final answer, use Continue to resume it instead of retrying from scratch.

Can users see the exact quota amount?

No. Users see percentages remaining for the daily and weekly allowances, plus their reset times. Admins can see exact operational usage for management.

Does quota differ by user type?

All KMITL users receive the same daily allowance and default weekly usage quota unless an admin adjusts a specific account's weekly limit.