Skip to content

gitflarePublicPartly agent-ready50

cf-workflows-pipeline

Durable content pipeline on Cloudflare Workflows: ingest, extract, summarise with Workers AI via AI Gateway, human approval, publish.

Fix pipeline runs failing at ingest (steps.attempt NOT NULL)629b2fa
Markdown57 lines2.0 KB

cf-workflows-pipeline

Deploy to Cloudflare

A durable content pipeline in a "Mission control" UI. It shows Cloudflare Workflows (step.do with retries and timeouts, step.waitForEvent human approval), Workers AI through AI Gateway, a D1 mirror of step state, and a live step timeline.

Architecture

flowchart LR
  B[Browser] -->|POST /api/runs, poll 1s| W[Worker]
  W -->|create / sendEvent| P[ContentPipeline Workflow]
  P --> I[ingest] --> E[extract] --> S[summarise]
  S -->|env.AI.run via AI Gateway| AI[Workers AI]
  S --> A{{waitForEvent approval}}
  A -->|approve| U[publish]
  A -->|reject| R[rejected]
  P -. mirror state .-> D[(D1 runs, steps, published)]
  W --> D
  B -->|/p/id| W

See docs/ARCHITECTURE.md and docs/API.md.

Local dev

npm install
npx wrangler d1 migrations apply DB --local
npm run dev

Tests

npm test (Vitest, Node environment) and npm run typecheck.

Deploy

npm run deploy, npm run deploy:staging. Deploying with the button or Gitflare Deploys creates the D1 databases: wrangler.jsonc deliberately has no database_id values. If deploying by hand, create the databases and add their ids.

Configuration

NameKindPurpose
DBD1runs, steps, published
AIWorkers AIsummarisation
PIPELINEWorkflowContentPipeline
AI_GATEWAY_IDvarAI Gateway id, default gitflare

Costs

Workers, Workflows, D1 and Workers AI have free allowances, then usage-based billing. Each run makes one model call; if AI fails, a deterministic extractive summary is used and marked as fallback. See Cloudflare pricing for current rates.

Screenshots

Add screenshots of the timeline and approval card here.

Built with Gitflare

This repository was built by cooperating coding agents on Gitflare, verified with npx vitest run.

License

Apache-2.0