# Usage and quota

Workspace usage combines conversation model cost and non-conversation model work, including Analysis and configuration tasks recorded in the usage ledger. Owners and admins can view the Usage settings page. Members and API keys cannot read this management view.

Agent Testing also counts toward the same workspace allowance. Agent replies, simulated-user turns, test evaluators, LLM parameter matchers, and repeated-run failure bucketing record their own purposes in the usage ledger. The Test Run detail attributes tokens and cost to its invocation; the workspace Usage page includes that cost in the daily total.

## Read daily cost and workspace activation

The page shows today's **daily cost**, the configured daily limit, remaining allowance, and the next reset. The accounting day is calculated in UTC; the dashboard presents the reset in the viewer's local time. Cost can continue to include Analysis or other model operations after the conversation's visible reply.

A new workspace starts with a daily limit of `0`. That means the workspace is not activated for conversations and the runtime returns `workspace_not_activated`. Only an Operade operator can activate a workspace or set or increase its daily allowance. Workspace owners and admins can view usage but cannot change that allowance.

When combined usage reaches the positive limit, the runtime returns `quota_exceeded`. The Usage page applies to every agent in the workspace, not to one agent or branch.

## Check Knowledge storage usage

The Knowledge page reports **knowledge storage usage** as used bytes and the configured workspace storage limit. This is separate from daily dollar quota: document bytes govern whether more Knowledge can be added, while model cost governs conversation and model operations.

Removing Knowledge can free storage, but force deletion may detach current agent drafts or active configurations under the documented safeguards. Read [Knowledge limits and troubleshooting](/docs/knowledge/limits-and-troubleshooting) before deleting dependencies merely to make space.

## Understand when quota is enforced

Quota is checked when a new conversation opens and again before the next message in an open conversation. A reply already in progress is allowed to finish completely even if that work reaches the limit. The following message is rejected before its user message is recorded, and the open conversation closes with the quota end reason.

This distinction explains why one final reply can appear after the Usage card reaches its limit. It is deliberate: the runtime does not truncate a reply already being produced. New channel segments can also be rejected before a conversation row is created.

## Recover safely

For `workspace_not_activated`, confirm the limit is zero and coordinate with the Operade operator responsible for activation. For `quota_exceeded`, wait for the UTC reset or have that operator review the allowance. Publishing another version, switching branches, or creating another API key does not bypass a workspace-wide limit.

After recovery, start a new conversation and confirm its status under [Conversations](/docs/monitor/conversations). If a previously open conversation is already closed, it will not accept another turn; start a new one instead. Use [Analysis results](/docs/monitor/analysis-results) to identify post-conversation model cost.