Limits

Fixed limits in this version, in one place. Limits you can configure are in the Configuration tables.

Requests

Limit	Value
Request body	16,000,000 bytes
User message	100,000 characters
History	2,000 messages, 32 images
Attached images	8 per turn, 5 MB each; PNG, JPEG, GIF, WebP
X-Vessel-User	1 to 128 characters: letters, digits, . _ : @ -
Conversation id (contextId)	1 to 128 characters: letters, digits, . _ : -
Token sub	1 to 200 printable ASCII characters, no spaces
Token clock leeway	60 seconds

Tools

Limit	Value
Process output kept	1,000,000 bytes
Output in tool.finished events	4,000 characters
read	400 lines by default, 2,000 at most; lines cut at 4,000 characters
grep	100 results by default, 1,000 at most; 400 columns per line
glob	200 results by default, 2,000 at most
list	500 entries
webfetch	4 MB, 30 seconds, 5 redirects
websearch	8 results by default, 20 at most
Service answer	1,000,000 bytes
Open files per call	4,096

Files and components

Limit	Value
File size	4 MB
Files per call	4
Files per turn	8, 16 MB in total
File name	100 characters
Components per answer	12, 200,000 bytes of JSON

Model and timing

Limit	Value
Model retries	3, after 1, 2 and 4 seconds, on 429, 500, 502, 503, 504 or a failed connection
Model read timeout	180 seconds without data
Streamed text merging	One event per 50 ms or 512 characters
Wait for a cancelled turn to finish	15 seconds
Sandbox readiness checks	Every 5 seconds until ready, then every 60
Task store sweep	Hourly
Issuer key refresh	At most once a minute, when a token names an unknown key

Containers

Container	Memory	Processes	/tmp
vessel	1 GB	256	64 MB
egress	512 MB	128	read-only root, no /tmp
search	512 MB	128	32 MB
sandbox	memory (2 GB)	pids (512)	scratch (256 MB)

Egress proxy tunnels close after 16 MB or 60 seconds of silence.