Dicey Table
Docs
Search the docs

Reference

AI agent limits and quotas

Every limit on a DiceyTable AI agent in one place: call rates, keys and sessions, pairing codes, file sizes, URL imports, screenshots and play-test tools.

2 min readVerified

View as Markdown
On this page8 sections

DiceyTable puts a cap on how fast an agent can work and how big its files can be. The caps keep your tab smooth, keep the shared servers fair, and stop a confused agent from running away with your project. Every number is below, grouped by what it limits.

When an agent hits a limit, it gets an error code that names it, such as rate-limited or too-large. The code and its fix are in AI agent error codes. None of these limits depend on your plan. AI agent access is free on every plan.

How fast can an agent call tools?#

Your tab counts calls for each agent session. Each kind of call has its own allowance. A session may spend a burst at once, and then the allowance refills at a steady rate. A call over the limit answers rate-limited with the wait in milliseconds.

Kind of call Steady rate Burst
Reads 50 a second 100
Writes (edits in the editor, lobby and settings) 20 a second 60
Table actions (table_* changes, and any write on a table or preview tab) 15 a second 30
Screenshots, describe_panel and capture_surfaces 2 a second 4
Diagnostics tools 10 a second 20
Calls refused before they run (unknown tool, wrong tab, bad arguments) 5 a second 20

A batch call counts once for each thing it changes. For example, a fields_set with 10 values counts as 10 writes. One batch never costs more than the burst, so a legal batch always gets through in the end.

Keys, sessions and pairing#

Limit Value
Active access keys per account 10
Key lifetimes you can pick 7, 30, 90 (the default) or 365 days
Agent sessions connected at once, per account 5
Tabs paired with one session 16
Pairing code lifetime 5 minutes
Wrong tries per pairing code 5
Pairing ticket lifetime (the bridge renews it while it runs) 5 minutes
Bridge heartbeat every 30 seconds
A session with no heartbeat ends after about 90 seconds
Time to answer a confirmation dialog 120 seconds, then it counts as Deny
Agent name (--client-name) The bridge keeps 64 characters; the AI badge other players see shows at most 32

Pair a DiceyTable tab with your agent explains codes and tickets, and create an API key covers key lifetimes.

How big a file can an agent import?#

A file has two limits. One is how big it can be on the way in. The other is how big it can be once it is stored in your project, after any conversion.

Limit Value
A file read from your computer, or fetched from a URL 128 MB
A stored file in your project 50 MB
A stored .onnx AI model 64 MB
A stored .hdr sky 10 MB
A stored file you can still publish from the app 45 MB (bigger files save with a warning)

Import models, images and sounds with an agent walks through an import.

URL imports#

Limit Value
Longest URL 2,048 characters
Redirects followed 5
Time to connect / to the first byte / in total 10 s / 30 s / 120 s
Fetches per key 60 a minute
Bytes per key 5 GB in any 24 hours
Fetches running at once 2 per account, 4 on the whole server

How URL imports are fetched safely explains each outcome these limits can cause.

Screenshots#

Limit Value
Default long edge 1,280 px
Smallest and largest edge 16 px to 2,576 px
Default maxBytes 600,000 bytes
Largest maxBytes 5,000,000 bytes
Largest image over the relay about 400,000 bytes
Entities one capture can frame or crop to 200
Time to click Allow tab capture for exact capture 60 seconds
Quiet time after you decline exact capture 10 minutes
Exact capture stops by itself after 5 minutes unused

Connections#

Limit Value
Message size between a tab and the hub on your computer 8 MB
Message size over the relay 2 MB
Bridges using the relay at once, per account 5
Tabs using the relay at once, per account 20

How the bridge works explains the hub and the relay.

Play-test and diagnostics tools#

Limit Value
Assertions in one table_assert call 100
Steps in one table_step_physics call 1 to 600
table_wait_settled timeout 100 ms to 60 s (default 10 s)
Rows the diagnostics hub keeps 2,000 (the oldest drop first)
Rows in one diagnostics_get page 500
diagnostics_watch wait 30 s by default, 60 s at most

The diagnostics sources reference has the hub's other limits.

Did this page answer your question?