AI Usage & Pay-as-You-Go

Every paid plan includes monthly AI quotas, counted in requests: AI discoveries, AI chat requests, AI flow generations and managed agent runs. This page covers how to see what those requests consume, and how to let them continue once a quota is used up.

Quotas and what counts against them

Quota What consumes it
AI discoveries Repository discovery, and ADR or documentation imports that use AI
AI chat requests Questions in the architecture chat, including the Slack and Teams bots
AI flow generations Generating a flow with AI
Managed agent runs Each run, including continued runs and approved held runs

Insights, agent planning and memory, and architecture diffs don't consume a quota. Calls made with your own AI provider key never consume one either: see Bring Your Own AI Provider.

When a quota is used up, its features stop until the next billing period, unless the organization has turned on pay-as-you-go.

AI usage card

Organization > Subscription shows a live AI usage this period card, refreshed every 10 seconds while the page is open. It splits the period's tokens by who pays for them:

  • Included in your plan: tokens used by requests within your quotas. They cost nothing beyond your subscription.
  • Pay-as-you-go: tokens used beyond your quotas, with their amount in euros so far. When pay-as-you-go is off, this block says so and links to the setting.
  • On your own key: tokens on your organization's own AI provider key, never billed by Archyl.

Input and output tokens are always shown separately, since they are priced differently.

Below the totals:

  • A filter (All, Included, Pay-as-you-go, Own key) applies to everything under it.
  • A per-day chart with a token axis. Hovering a day shows its date, input, output and total tokens, plus the cost when the Pay-as-you-go filter is selected.
  • A breakdown by feature, with input and output tokens.

Tokens of included requests are shown for transparency only: your plan includes requests, not a number of tokens.

Usage statement (CSV)

At the bottom of the card, Statement (CSV) downloads the usage of the current or previous billing period, one row per day, feature and billing type:

date,feature,billed_as,calls,input_tokens,output_tokens,cost_eur
2026-10-22,chat,included,212,1268000,171000,
2026-10-23,chat,pay_as_you_go,48,288000,38400,0.119232
Column Meaning
date Day, UTC
feature The AI feature, for example chat, discovery, file (discovery reading files), managed_agent, flow_gen
billed_as included, pay_as_you_go or own_key
calls Number of AI calls
input_tokens, output_tokens Tokens read and written
cost_eur Amount billed, for pay_as_you_go rows only

Any organization member can see the card and download the statement.

Pay-as-you-go

With pay-as-you-go on, a request that would be refused because its quota is used up runs anyway, and only the tokens it uses are billed, on your next invoice. Requests within your quotas stay included.

Availability

  • Business and Scale plans, with an active subscription paid by card. Not available on the free Developer plan, or on custom contracts invoiced outside the app.
  • Only the organization owner can turn it on, change the cap or turn it off. Other members see the setting read-only.
  • Off by default.

Prices

Usage Input Output
AI requests (chat, discovery, flow generation, agents) €0.23 per 1M tokens €1.38 per 1M tokens

Prices exclude VAT. A request is priced when it is made: if prices change later, past usage keeps the price it was billed at. As an order of magnitude, a typical chat answer reads about 6,000 tokens and writes about 800, roughly €0.0025.

The card in Organization > Subscription always shows the current prices.

Turn it on

  1. Go to Organization > Subscription and find the AI pay-as-you-go card.
  2. Tick Continue beyond quotas, billed per token.
  3. Set a monthly cap, between €5 and €10,000. It defaults to €50.
  4. Tick the agreement to be billed per token. Save stays disabled until you do.
  5. Click Save.

If a quota runs out before it's set up, the quota message offers a shortcut to this card.

To turn it off, untick the setting and save. Requests past a used-up quota are refused again straight away.

Monthly cap

The cap bounds what pay-as-you-go can cost in one billing period. The card shows what has been spent against it. Once spending reaches the cap, requests beyond your quotas are refused until the next period, or until the owner raises the cap.

Requests already running when the cap is reached complete and are billed, so the final amount can pass the cap by the cost of those requests.

On your invoice

Pay-as-you-go appears as its own line on the monthly invoice, AI pay-as-you-go, in units of €0.0001. The invoice memo adds a summary of the tokens behind it, so the amount can be checked without opening Archyl:

AI pay-as-you-go: AI usage beyond the plan quotas, 1 Oct 2026 to 1 Nov 2026.
- AI requests: 6,920,000 input tokens x EUR 0.23/1M + 812,000 output tokens x EUR 1.38/1M = EUR 2.7122
Billed: 27,122 units x EUR 0.0001 = EUR 2.7122 excl. VAT (each report to billing is rounded to EUR 0.0001).

Usage is reported to billing every few minutes, and each report is rounded to €0.0001. For the day-by-day detail, download the statement.

The summary is added to the regular end-of-period invoice. Invoices issued immediately by a change in the middle of a period, such as a plan change or a cancellation, carry the line without the summary; the statement has the detail.

FAQ

Does turning on pay-as-you-go change what's included? No. Requests within your quotas stay included. Only usage beyond a used-up quota is billed.

We use our own AI provider key. Does this apply? No. Calls on your own key don't consume quotas and are never billed by Archyl.

Can a member run up costs? Only within the cap the owner set, and only once the owner has turned pay-as-you-go on.

Why do discovery and agent runs cost more than a chat answer? They read much more: whole files, the architecture model, the repository. The usage card and the statement show what each feature actually cost.