1. Automation Jobs & Execution Engine

Automations are now backed by a unified execution engine that makes sure every upload creates the right jobs and that decisions are executed in a predictable way.

• Triple-layer safety for automation jobs: creation on document type create/update and a final fallback on upload to guarantee jobs are never silently skipped.
• New backend engine in lib/automation/engine.tsand lib/automation/executor.ts to turn LLM-produced automation descriptions into concrete decisions and actions.
• Built‑in actions for sending emails and calling external HTTP APIs, with robust error handling and structured payloads containing job/document context.
• Manual approval support on a per‑rule basis: rules can require approval before execution, and decisions are tracked on the automation job record.

2. Agent Edit UX & Automation Sync

The agent edit experience has been simplified and made consistent with creation, while ensuring automation workflows stay in sync across UI surfaces.

• Edit page for processing agents now uses a single‑step layout that matches the create page – no more multi‑step wizard. Workflow configuration lives inline with agent settings.
• Automation loading always reads from the automations table first, falling back to document_types.automation only when needed, so table‑based edits and edit‑page edits stay aligned.
• Saving an agent now keeps document_types.automation andautomations.json_data in sync to avoid stale configurations.

3. Smart Document Uploader – No-Agent Handling

When no suitable processing agent exists, uploads no longer fail silently. Instead, the system guides users to create the right agent while safely preserving their documents.

• Smart uploader can detect when no agents exist or when GPT‑powered routing cannot find a match, and responds with a needsAgentCreation flag.
• Documents are stored under a pending location and shown in the UI with a dedicatedwaiting_for_agent status instead of being lost.
• A "Create Processing Agent" modal lets users quickly define a new agent; once created, the pending document is automatically queued and processed.
• Create‑page integration surfaces pending documents, so advanced configuration flows can still auto‑process the original upload when finished.

4. System Agent Templates

New system templates give every workspace a set of high‑quality starter agents without any manual setup.

• New system_agent_templates table with pre‑built agents (invoice processor, CV parser, receipt scanner, contract analyzer, and more).
• Automatic duplication of system agents on signup or when an existing user has no agents, wired through /api/system-agents/duplicate and the signup service.
• Dashboard agent list shows a "Pre‑made" badge for system‑derived agents so users can easily distinguish templates from custom agents.

5. PDF Experience & Streaming Improvements

PDF handling has been upgraded for both previewing and multi‑page AI streaming.

• New modal PDF preview component with zoom, page navigation, rotation, download, fullscreen, and keyboard shortcuts, built on react-pdf.
• Dedicated hook for opening and controlling previews from anywhere in the app.
• Multi‑page AI streaming now correctly auto‑advances through pages, keeps streamed text cached per page, and restores content when navigating between pages.

6. Compatibility & Migration Notes

• Existing document types and automations continue to work; missing automations are backfilled automatically on create/update/upload.
• System agents are additive – they appear as normal agents that users can fully customize without affecting the underlying templates.
• No manual data migrations are required for this version; all fallbacks and sync logic are handled in application code.