Pipelines and tasks

Workflow catalog

The platform has typed workflows for repeatable operations and a generic task path for exceptional one-off work.

Pipelines

Pipelines are guided launch points for repeatable AI work. A launch form validates workflow-specific configuration, combines it with the active organization and venture, resolves the assigned agent and model, then starts a durable run.

Every dispatch carries an organization ID and venture ID. Repository, department, goal, resource-context, and per-launch model settings are added only when required by that workflow. The server performs authorization; client-supplied IDs are not treated as proof of access.

Unified work feed

Tasks provides a combined history of workflow runs and development work. Use filters to narrow by status, workflow type, or time. Open a row to inspect details, phase progress, output, approvals, errors, and costs.

Common states include queued, running, awaiting input, completed, failed, canceled, and timed out.

Durable execution

Long-running workflows use durable orchestration, so run state and retry history survive application restarts. The workflow detail provides current phase, status, output, event timeline, model usage, cost, and external delivery references.

Depending on the workflow and state, users or operators can:

• submit requested input;
• approve or reject a decision;
• cancel active work;
• retry a failed activity or workflow;
• recover work interrupted by infrastructure;
• inspect execution and cost details.

Costs and status updates

Workflow pages report model usage and phase costs when providers return usage data. Usage is attributed to an organization, venture, run, workflow type, phase, model, and provider where those dimensions are available. A missing cost is not proof that a phase did not execute; provider usage and application persistence are separate failure domains.

Live state uses polling or streamed events depending on the surface. A short delay can occur between execution and the visible event/cost update.

Run identity and observability

Capture these fields before escalating a workflow problem:

Field	Why it matters
Run or task ID	Locates the application record.
Workflow/run ID	Locates durable execution history.
Workflow type and current phase	Selects the correct state machine and recovery action.
Organization and venture	Confirms tenant scope.
Agent, model, and provider	Isolates execution and credential failures.
Last event and timestamp	Distinguishes slow work from stalled work.
Error type and retry count	Shows whether retry is automatic, safe manually, or exhausted.

Older links and records

Older Pipelines and Content Review URLs may redirect to their current venture-scoped destinations. Bookmarks should be updated to the new location.

Older development records can expose fewer details than current workflow runs.