Limits and allowances

This page lists each plan’s allowances, what counts against them and the size limits that apply to every request. Rate limits set how fast you can send requests.

Allowances by plan

Plan	Price per month	Reads per month	Writes per month	Stored memories	Projects
Free	$0	10,000	1,000	500	3
Starter	$20	50,000	5,000	5,000	10
Growth	$99	300,000	30,000	50,000	Unlimited
Scale	$499	3,000,000	250,000	500,000	Unlimited

Allowances apply to the whole account, across all its projects. Monthly allowances reset at 00:00 UTC on the first day of each month. Custom plans have allowances set for your workloads.

What counts

Reads

Request	Reads used
Context: `POST /v1/memories/context`	5
Inspect: `GET /v1/memories/{memory_id}/inspect`	3
Recall, search and get	1
Each page of a list: `GET /v1/memories`	1
MCP `context` and `inspect_memory`	5 and 3
Other MCP read tools and the `alvia://memory/{memory_id}` resource	1

A read counts when it completes, whatever it returns: a context or inspect 404 uses its reads too. A read refused with 400 for bad input, or that fails with a 5xx, uses none. A read is refused when fewer reads are left than it costs, so a context call needs five.

Writes

Request	Writes used
`POST /v1/memories` with `infer: false`	1
`POST /v1/memories` with `infer: true` (the default)	1, plus 1 for each fact formed from the passage
`POST /v1/memories/import`	1 for each chunk of the conversation, plus 1 for each fact formed from it
MCP `remember`	The same as `POST /v1/memories`
Forget and feedback	0
A repeat sent while the first copy is still forming	0

Every accepted write counts, whether it forms new memories, matches existing ones or forms none, except a repeat sent while the first copy is still forming.

Stored memories

Each active memory counts once, however many versions it has. A forgotten memory stops counting. The stored-memory limit does not reset each month.

Demo Project and Playground

Every new account starts with a Demo Project of sample memories. It does not count toward the project limit, and its sample memories do not count toward the stored-memory limit. Memories you add to it count as usual.

Calls you run from a project’s Playground in the dashboard count like any other call: List, Search, Recall and Context use reads, and Create uses writes.

When you reach an allowance

Error	Cause
`429` `rate_limit_exceeded` `monthly_read_quota_reached`	The monthly read allowance is used up. Reads are refused until the allowance resets or you upgrade.
`429` `rate_limit_exceeded` `monthly_write_quota_reached`	The monthly write allowance is used up. Writes are refused until the allowance resets or you upgrade.
`409` `conflict` `memory_limit_reached`	The account holds its plan’s stored-memory limit. New writes are refused until you forget memories or upgrade.

This applies on every plan. There are no overage charges.

A monthly 429 refusal carries limit, used and resets_at, and a Retry-After header with the seconds until the reset:

{
  "type": "https://alvia.ai/docs/errors/#rate_limit_exceeded",
  "title": "Rate limit exceeded",
  "status": 429,
  "detail": "This account has used its monthly write allowance.",
  "instance": "/v1/memories",
  "code": "rate_limit_exceeded",
  "reason": "monthly_write_quota_reached",
  "limit": 1000,
  "used": 1000,
  "resets_at": "2026-11-01T00:00:00Z"
}

A passage or import that reaches the write allowance partway forms what fits in the remaining allowance, in order, and returns 202. The response does not say what was left out. Check the remaining allowance in the dashboard before a large import. See Cost in Import conversations.
Forget and feedback are never refused by an allowance, so you can always free stored memories.

Size limits

Item	Limit
Request body	5 MiB
`content` on `POST /v1/memories`	1 to 32,000 characters
`content` with `infer: false`	1 to 1,000 characters
`query` on context, recall and search	1 to 1,000 characters
`limit` on search	1 to 50, default 25; values outside are clamped
`limit` on list	1 to 20, default 10
Each scope field	128 characters
Messages in one import	1 to 500
Each imported message	16,000 characters
All messages in one import	320,000 characters
Characters without a space in a message	2,000
`speaker` on an imported message	128 characters

Character counts are UTF-16 code units, as in JavaScript’s string.length. Some characters, such as emoji, count as two.

Search’s limit is clamped to its range. A request over another size limit gets a 400 response with code validation_error, a 400 with code invalid_argument for a conversation too large to process, or a 413 response when the body is over 5 MiB.