Limits

The numbers the gateway works within, in one place.

Requests

Limit	Value	What happens past it
Time to the first byte of the provider's answer	60 seconds	504 upstream_timeout. Not retried on another destination.
Whole request, streaming included	600 seconds	The request is cut off.
Where model must appear in the body	Within the first 256 KiB	400 model_not_determinable. Send the name in X-Onehouse-Model instead.
Request body that can be retried on another destination	8 MiB	A larger body goes to one destination only.
Attempts per request	2	The first destination, then at most one more. See Routing and failover.
Skip after a provider's 429	Its Retry-After, or 10 seconds; at most 60 seconds	The destination is skipped for that model.

Tokens

Token	Lasts
A person's gateway token	7 days
A service principal's gateway token	5 minutes

See Sign-in and tokens.

Names and setup

Limit	Value
Alias name	Lowercase letters, digits, dots, dashes and underscores; up to 64 characters
Model description	Up to 512 characters
Models published at once from Add a model	100
MCP server name	Lowercase letters, digits and hyphens; up to 64 characters
MCP server request timeout	1 to 300 seconds
MCP transport	Streamable HTTP only
People in Revoked user access	10,000

Usage

Limit	Value
How long usage is kept	30 days
Rows in each Usage table	100. Totals still count every row.
Callers attributed by name	500 distinct callers, or 10,000 caller-and-model pairs. Past that, usage is folded into Beyond attribution cap.

See Usage.

By cloud

Capability	AWS	Google Cloud
AWS Bedrock as a provider	Yes	No
Direct credential (a key Onehouse stores) for providers and MCP servers	Yes	No
OAuth MCP servers	Yes	No
Onehouse MCP server	Yes	No