Automation without code
You already automate parts of your document workflow: a person watches a folder for invoices, merges them, runs OCR, and emails the CFO. That's a pipeline — just a slow, human-powered one.
Pipelines turns it into a versioned graph.
How to build one
- Open /pipelines/builder
- Drag nodes onto the canvas — each is a PDF tool or AI action
- Connect them with edges — output of one becomes input of the next
- Pick a trigger: cron schedule, webhook, or manual run
- Save and activate
Your pipeline is stored as JSON — nodes, edges, trigger_config — and executed by a Celery worker that does topological sort and runs each node with the previous node's output.
What's in the toolbox
Every PDF tool is a node:
- Input: uploaded files, URL download, Google Drive folder watch
- Process: merge, split, compress, OCR, redact, translate, extract
- Output: download, upload to S3, post to Slack, email, webhook
Why versioning matters
The pipeline JSON is stored immutably per version. When you edit:
- The existing live version keeps running for scheduled triggers
- Your edits create a new revision
- Activate the new one, and the cron switches over
- Deactivate to roll back in one click
Use cases we've seen
- Daily invoice ingestion — Drive folder → OCR → extract totals → Google Sheets
- Contract redline pipeline — upload → Smart Redact → compare-explain vs template → Slack alert
- Academic paper translation — PDF → OCR → translate to target language → save as new PDF
The limits
- Each pipeline can have up to 50 nodes
- Max execution time: 10 minutes per run (raise via Business plan)
- No branching based on content (yet — conditional edges are next quarter)
Active on all paid plans. Start at /pipelines.