Agentic Apps, Job Retry Resilience, and Rate-Limit Visibility

Features

• Agentic Apps: new Apps page lets you create and manage agent-backed applications, with per-app agent assignment and launch controls.
• Provider queue-depth SSE: provider queue and active job counts now update in real time via Server-Sent Events, so the dashboard always reflects live load without a page refresh.
• Rate-limit visibility: when a provider hits a usage or rate limit, the dashboard surfaces the status and estimated reset time directly on the provider card.
• Job retry on transient failure: jobs that fail due to a provider error (rate limit, worker crash, etc.) now stay in the Redis stream PEL and are automatically re-dispatched after exponential backoff, rather than being permanently failed.
• Credential editing for team admins: team admins, team owners, and org admins can now edit existing credentials, not just org owners.

Improvements

• Activity feed simplified to two events per job — “Job submitted” when the job row is created and “Job completed / Job failed” when it finishes. Retry and re-dispatch events no longer generate duplicate activity entries.
• Large clipboard pastes now render as collapsible chips in the chat input rather than raw text blocks, with the content visible in the debug panel.

Fixes

• Fixed image vision token handling and tool-continuation job wiring for chat turns that include images or multi-step tool use.
• Fixed stale TypeScript import (AppEmptyState) that was breaking the production web build.