Limits
Corlecti Map meters one thing: the model's own work. Everything else on a map — writing, connecting, searching, exporting, sharing — spends nothing, on any plan.
The AI allowance
You have an allowance, not a balance: two windows, one daily and one weekly, each shown in the app as a percentage used and a date it resets. There is no monthly window — nothing here waits longer than a week.
AI work, relative to Free day week
Free base base
Pro ×5 ×5
Workspace ×5 ×5The weekly window is less than seven dailies, on purpose: it exists so a bad week cannot be spent in one sitting, not so a single heavy evening waits until Monday. The daily window resets every day regardless of the week's state, so the longest pause after one heavy evening is until tomorrow.
What spends it:
- The agent building or editing a map, and a question put to it in chat.
- Importing a PDF (up to 15 MB) or a linked page (up to 5 MB), together with
building the structure out of it.
- The
delegatetool over MCP — the one MCP tool that runs a
model on Corlecti Map's own side rather than in your client.
- Buying one more map slot once your plan's included maps are used — priced in
the allowance itself (a whole weekly window), never in money, and yours permanently once bought.
- Coming: a map on a schedule, edited by the agent while you are away.
What spends nothing, on every plan: manual work on the canvas, the Telegram bot and the browser clipboard, the graph lens and every finding in it, both kinds of search, export in any form, publishing a map, version snapshots, dictation in the browser, and anything the agent does when your own provider key is connected — see "Bring your own key" below on this page.
When the allowance runs out, only the model stops. The agent declines to build or edit and says when the allowance resets; the canvas, search, export and the bot all keep working exactly as before.
Voice transcription — a separate window
Recognising speech (dictation in the browser, a voice note to your Telegram bot) is billed by Corlecti Map, but never against the AI allowance above — it has its own daily and weekly windows, doubling rather than growing fivefold on paid plans:
Voice transcription, relative to Free day week
Free base base
Pro ×2 ×2
Workspace ×2 ×2MCP calls — a volume quota, not the allowance
Every call to the MCP server is counted, on a quota that is separate from the AI allowance above and never converted into it — calling a tool costs nothing from your model budget, even on a heavy session, unless the tool is delegate.
MCP calls, per plan Free Pro Workspace
Per day 100 500 1000
Per week 500 2500 5000
Live keys at once 5 25 50A refusal names the count and when it resets, never a percentage — this is a different meter from the allowance, and mixing their units would answer two questions with one number:
Corlecti Map MCP quota reached: 100 of 100 calls today. Resets 2026-09-11 00:00 UTC. The daily and weekly quotas are separate.
Rate limits — a short burst ceiling, separate again
Beneath the volume quota above, every paid surface also has a short-window ceiling against runaway loops and key-guessing, set well above ordinary use — you should not meet it working normally:
Surface Limit Window
Chat (a turn) 20 5 minutes
Search 30 1 minute
Voice transcription 10 1 minute
MCP (`/api/mcp`) 300 1 minuteA 429 names how many seconds to wait and never touches your plan:
{ "error": "Too many requests. Try again in 12 second(s).", "code": "rate_limited", "retryAfter": 12 }Retrying immediately will not help; wait the stated number of seconds.
Refusal codes
- 401 — the key is missing, malformed, or has been revoked. Mint a new one
in Settings → MCP Server and replace it.
- 402 — a capability or a spend gate, never a rate limit. The body says
which:
- Your account has used today's, or this week's, AI allowance — wait for the
date it names, or connect your own provider key to keep working unmetered.
- A feature named in the body is not included on your current plan.
- You have used every map slot on your plan — buy one more (a weekly
allowance window, not money) or free one up.
- You opted into extra usage beyond your plan's allowance and either your
own monthly cap on it, or your Corlecti account balance, is now reached — raise the cap, add balance, or wait for your plan's own allowance to reset.
- 403 — the one case that is not a 404: a workspace member scoped to one
branch reached outside it. Rare outside shared maps, and the body names how many blocks were refused, never which ones.
- 404 — the map does not exist or is not yours. Deliberately the same
answer either way, so a wrong id cannot be used to learn that somebody else's map exists.
- 429 — see "Rate limits" above on this page.
- 500 — ours. If it repeats, [email protected].
- 503 — a dependency this deploy needs is unreachable (account status, or
the file store for an image or a recording). Not your plan's fault; try again shortly.
A tool call that fails for an ordinary reason — a map that is not there, a document that will not validate — does not use a status code at all: the call succeeds and the result is marked as an error, with the reason inside it. See the API page "What an error means".
Plans at a glance
Figures only — pricing is on the app's own upgrade page.
Free Pro Workspace
Maps included 10 50 100
Files, pictures and recordings 100 MB 1 GB 2 GB
Board tabs per map 5 12 30
Version snapshots (named) 20 20 20
Own Telegram bots 1 5 unlimited
Guests in your own bot — 5 5An included map is a ceiling, not a wall: past it, one more map costs a whole weekly AI allowance window rather than money, and it is yours for good — see "The AI allowance" above on this page. Blocks and connections are never limited, on any plan.
Coming:
- Scheduled maps at once — 2 on Free, 10 on Pro, 25 on Workspace.
- A key narrowed to a folder, or seated on one branch — see
the MCP page's "Narrowing a key" section.
- Notifications delivered outward by email or by Telegram — see
"Notifications" below on this page.
Bring your own key (BYOK)
Connect your own key to a model provider — Anthropic, OpenAI, DeepSeek or Google Gemini — in Settings → AI Chat Model, on any plan including Free. Everything the agent does then runs on that key, and spends none of your Corlecti Map allowance at all: the choice is per account, and reversible at any time.
Your key is encrypted at rest with AES-256-GCM before it is stored, decrypted only for the moment a request to your provider is made, and never returned by any tool, route, or reply — not even to you. Settings → AI Chat Model shows only its last four characters.
Notifications
The in-app inbox — an edit on a shared map, a reply in its conversation, a comment — spends nothing on every plan, and works today. Coming: a row for a scheduled run finishing, once schedules themselves run (see "Plans at a glance" above).
Coming: delivery outward, to email and to Telegram, on Pro and Workspace. Both channels will ship off by default — turning one on for yourself will be your own choice, never a default. Delivery will be a digest, never one message per event: one send per person per window, naming only what is still unread, and opening the app will cancel a queued email — the single best proxy this product has for "the person already knows."