Limits

The values below are the values on proxium.tech today. A limit of 0 means no limit.

Request and response size

Limit	Value	At the limit	Source
Request body	32 MiB	`413`, with a plain-text body.	`src/server.rs::max_body_bytes`
Provider answer, buffered	64 MiB	The attempt fails. proxium moves to the next entry of the plan.	`src/upstream.rs::read_capped`
Stored answer body	32 K characters, where 1 K is 1,024	proxium cuts the stored copy. The caller gets the full answer.	`src/types.rs::BODY_MAX_CHARS`
Model id on image, speech and video calls	200 characters of `A-Z a-z 0-9 . _ - :`	`400 bad_model`.	`src/server.rs::model_id_is_urlsafe`

Limit	Value	At the limit	Source
`x-proxium-timeout-ms`	At most 120000 ms	proxium lowers a larger value to this value. It ignores 0, a negative value and text. The next row uses the lowered value.	`src/engine.rs::deadline_from_headers`
Whole failover plan, with `x-proxium-timeout-ms`	The header value	`502 upstream_failed`, with `client deadline exceeded` in the message.	`src/engine.rs`
Whole failover plan, without the header	No limit	Each attempt keeps its own limit.	`src/engine.rs`
One buffered attempt, not streamed	120 s	The attempt fails as a timeout. proxium moves to the next entry.	`src/upstream.rs::UpstreamClient::new`
Connection to a provider	10 s	The attempt fails. proxium moves to the next entry.	`src/upstream.rs`
Gap between two reads of a stream	300 s	proxium ends the stream. The timer also runs while the provider prepares its first byte. A stream has no limit on its total time.	`src/upstream.rs::stream_read_timeout`

With x-proxium-timeout-ms, the deadline covers the plan up to the first byte of a stream. After the first byte, only the gap between reads applies.

Limit	Value	At the limit	Source
Entries in one failover plan	4	proxium stops and answers `502 upstream_failed`.	`src/chain.rs::MAX_FAILOVER_CHAIN`
Extra tries of one entry	1	Only after a 5xx, 408, 429 or "warming" answer.	`src/engine.rs::EngineConfig`
Wait for a provider's `Retry-After`	30 s at most	proxium waits the smaller of `Retry-After` and 30 s, then tries again.	`src/engine.rs::EngineConfig`
Failures that open a circuit breaker	5 inside 10 minutes	proxium skips the provider.	`src/circuit.rs`
Open time after a transient failure	30 s	proxium allows a trial call.	`src/circuit.rs::TRANSIENT_RESET_MS`
Open time after a rejected credential or an exhausted quota	30 minutes	proxium allows a trial call.	`src/circuit.rs::QUOTA_RESET_MS`
Trial successes that close a breaker	2	The breaker closes.	`src/circuit.rs::PROBE_SUCCESSES`
Failover price guard	Not active	Failover can reach a model at any price.	`src/engine.rs::price_guard`

Limit	Value	At the limit	Source
Key calls per hour	0 on a key that the console creates	`429 tenant_capped`, `retry-after: 60`.	`src/accounts.rs::mint_team_key`
Key cost per day	0 on a key that the console creates. Not enforced on proxium.tech.	None.	`src/server.rs::budget_limits`
Key requests and tokens per minute	0 on a key that the console creates	`429 tenant_capped`, `retry-after: 60`.	`src/store_pg.rs::PgKeyStore::resolve`
Sender calls per hour	You set it. 0 by default.	`429 source_capped`, `retry-after: 60`.	`src/source_limits.rs`
Sender dollars per day	You set it. 0 by default.	`429 source_capped`, `retry-after: 60`.	`src/source_limits.rs`
Monthly cap of the project	None during the open beta	None.	`src/credits.rs`

The calls window is a rolling 60 minutes. The cost window is a rolling 24 hours, in hourly buckets. The rate windows are a rolling 60 seconds.

Limit	Value	At the limit	Source
Key name on the Keys screen	60 characters	proxium drops a longer name. The key is unnamed.	`src/accounts.rs::mint_team_key`
Endpoint name	60 characters	`400 bad_name`.	`src/tenant_api_endpoints.rs::create`
Vendor name on the Providers screen	1 to 40 characters of `a-z 0-9 -`, with no `-` at either end	`400 bad_name`.	`src/tenant_endpoints.rs::MAX_NAME`
Models of one vendor	200	`400 too_many_models`.	`src/tenant_endpoints.rs::MAX_MODELS`

Change	Value	Source
A routing, sender-ceiling or cache-charge change in the console	At once on the server that saved it, and within 30 s on every server	`src/main.rs::spawn_config_refresh`
The monthly spend that the credit check reads	Refreshed every 30 s	`src/main.rs::spawn_config_refresh`
The model list of each vendor	Read again every 21600 s	`src/model_refresh.rs::DEFAULT_REFRESH_SECS`