Project and site administration

Manage configuration and access, and operate optional AI within existing limits.

Open the right settings

Project settings in the sidebar manages the selected project's configuration. The cog beside the top-right avatar opens administrator-only Site settings. Appearance/email/session preferences are in the avatar menu, not the project role label.

Administrators create projects and manage accounts/invitations. Use the invited Google email exactly; an existing signed-in Google account does not automatically become a member of every project.

Membership and configuration

Managers configure project members and roles. Suspend or revoke access when appropriate; current checks apply to search, files, dashboards, notifications and integrations. Viewer access remains read-only for project content while permitting personal favourites.

Configure ticket types within epic/standard/subtask hierarchy levels and typed custom fields. Archive obsolete choices/configuration to preserve history. Workflow mappings on types apply to new tickets; existing tickets retain their workflow assignment.

Review board filters, published workflows and project release-readiness defaults after changing configuration. Planned releases copy defaults and may diverge; changing a project default does not silently rewrite every release.

AI controls

Open Site settings → AI & automation. Administrators control AI enablement, enabled models, the default model and usage limits. Project automation has separate enablement. Managers choose among enabled models in supported drafting/configuration controls.

Keep AI opt-in. Native Workers AI must satisfy the existing free-only verification gate; if that cannot be verified, leave AI disabled with its explanation. Local reservations and per-day limits are application controls, not the Cloudflare account's remaining allowance. The account-level free allowance may also be used by other services.

Model availability and provider pricing can change. Use the source links and verification date shown in settings; estimates are not billing guarantees. Consult Cloudflare Workers AI pricing before operational decisions. These guides do not authorize upgrades, credits or paid inference.

Automation operations

Use Project settings → Automation runs to inspect queued, running, approval, paused, failed and budget-blocked jobs. Manage them using the offered pause/resume/retry/cancel controls, and distinguish committed effects from pending work.

Before enabling a chain, confirm its publisher retains the necessary project rights, reviewers are eligible, template versions are published and scope is appropriate. Disabling capability or revoking access can block further actions; it does not retroactively undo previous commits.

Usage diagnostics

Expand Usage and background work in Site settings → AI & automation, then choose Refresh usage. Today's attempt outcomes distinguish completed calls, timeouts and provider failures. The measured-usage column says how many calls returned token counts; calls without provider counts remain unknown.

Nearer reserves UTF-8 source bytes plus 4,000 units per admitted AI attempt. These are conservative local units, not measured model tokens or Cloudflare neurons. Existing limits still apply. A timeout retains its reservation because the provider may have processed the request. Internal admission blocks and Cloudflare quota/rate-limit failures have different explanations; retrying does not expand either allowance.

The same disclosure shows queued work and the latest maintenance outcomes. Indexing, file cleanup, dispatch, retention, email and session cleanup run independently; a failure in one does not prevent the others. A failed or deferred task is retried by subsequent scheduled maintenance. Categories share a single invocation budget and take turns receiving first priority, so cleanup still progresses under sustained load.

Storage and background budgets

Attachments use private storage with a 10 MB per-file limit and configured total/operation budgets. Removed objects count until confirmed purge. Email and AI have independent bounded-operation controls. Default allocation is not a promise of free provider capacity for every workload.

Do not interpret a budget error as permission to enable a paid product. Use smaller imports, bounded reports and deliberate retries. Boards check for changes every two minutes while visible, avoid returning unchanged ticket snapshots and pause checks in hidden tabs. Dashboards refresh manually.

Operating checklists

For a new project: configure membership, types/fields, workflow mapping, boards, readiness defaults and shared queries; test as the intended roles before publishing process changes.

For a departing person: review membership/suspension, sessions, assigned work, published automation responsibility and pending reviews. Preserve history rather than rewriting authorship.

For recovery: follow backup restoration, which preserves local authentication and integration settings and requires explicit destructive confirmation.

Native background execution uses short resumable chunks and a conservative local allocation of 2,000 Workflow steps per UTC day. This leaves headroom within Cloudflare's currently documented 3,000 free steps/day but cannot account for other applications using the shared account. Allocations may exceed actual steps because unused reservations are retained. When this allocation is exhausted, queued work remains recoverable; it does not enable paid execution. Scheduled dispatch recovers work that could not be woken within an interactive request's SQL budget.