Limits and errors

Nenavi pays for the models that extract entities and claims from your text, so every project has hard caps. These are the alpha defaults. They can change, and the operator can raise the limits of one project.

Limits

Limit	Default	What counts	Error
Requests per minute	60 per key	Every request the service accepts	`429` `rate_limited`, with `Retry-After`
Content size	32 KiB	The `content` of one store or ingest	`413` `content_too_large`
Ingest calls per month	20,000	Ingests, fast ingests, stores that extract, each memory a re-ingest takes, and organize and extraction re-runs	`429` `monthly_ingest_calls_exceeded`
Content tokens per month	2,000,000	The tokens in the `content` of stores and ingests	`429` `monthly_content_tokens_exceeded`
Memories per project	50,000	Stored memories, plus saved text that is not indexed yet	`429` `max_nodes_exceeded`

Months are UTC calendar months.

In the console, an organization can have up to 10 projects, and a project can have up to 5 active keys. A deleted project still counts toward the 10 until its seven day grace period ends. Reaching either limit returns 409 limit_exceeded from the console’s API.

How the counts work

A store extracts by default, so it counts as one store call and one ingest call. Send "extract": false to store text with no extraction. Then it counts as a store call and its tokens only.
A fast ingest batch counts as one ingest call, however many items it holds. It adds the tokens of every item.
A retry of a fast ingest batch with the same Idempotency-Key and the same items costs one request. It does not count again toward the monthly limits.
A request that the service refuses before doing any work is not counted. Neither is a request that the engine answers with a 4xx error.
The check happens before the work. A request is refused when it would take the month past the limit, so you can reach a limit exactly.
The counts are close, not exact. With several servers, a project can pass a monthly limit by about one minute of traffic. The rate limit is also kept per server, so the effective limit can be higher than 60 per minute.

See your usage

The console’s Usage tab shows this month’s requests, store, ingest, retrieve, and MCP calls, and content tokens, with a bar against the two monthly limits. The same numbers are at GET /v1/usage, which needs a key. It takes no engine slot, so it is cheap to poll. It counts as one request.

The number of memories is at GET /v1/stats.

Errors

Every error from the hosted layer has a detail and a reason_code. Errors that the engine raises keep the engine’s { "detail": "..." } shape.

Status	`reason_code`	What to do
`401`	`api_key_missing`	Send `Authorization: Bearer <key>`.
`401`	`api_key_invalid`	Check the key. Copy it again from where you saved it.
`401`	`api_key_revoked`	Create a new key in the console.
`401`	`project_deleted`	The project was deleted.
`413`	`content_too_large`	Send less than 32 KiB in one `content`. Split the text.
`413`	`body_too_large`	Send a smaller body. The limit is 256 KiB, or 4 MiB on batch routes.
`422`	`content_invalid`	The text cannot be encoded as UTF-8.
`422`	`wait_for_extraction_unsupported`	Remove `wait_for_extraction`. Poll the extraction status instead.
`422`	`privacy_policy_unsupported`	Remove the `policy`.
`422`	`reembed_unsupported`	This route is off in the alpha.
`422`	`limit_out_of_range`	Change `limit` to a value in the range on The API.
`429`	`rate_limited`	Wait the number of seconds in `Retry-After`.
`429`	`monthly_ingest_calls_exceeded`	Wait for the next month, or send `"extract": false` to store without extraction.
`429`	`monthly_content_tokens_exceeded`	Wait for the next month.
`429`	`max_nodes_exceeded`	Archive memories you no longer need. The count is refreshed about once a minute.
`409`	`reingest_running`	A re-ingest for this user is already working. The `detail` names it.
`503`	`extraction_disabled`	Extraction is off on the service. Fast ingest still works.
`503`	`database_unavailable`	Wait a moment and send the request again.
`503`	`lease_unavailable`	The service was busy. Wait a moment and send the request again.

To raise a limit, contact the maintainer. The address is on the privacy page .