gem-hunter

AI workflows in Supabase, no extra infra.
TypeScript workflows with full autocomplete, zero boilerplate, automatic retries and realtime progress. Built on Postgres + Edge Functions.

Docs | Demo | GitHub | Discord

"A workflow engine built on Supabase primitives."
Paul Copplestone, CEO, Supabase (via X)

# Install pgflow in your Supabase project
npx pgflow@latest install

# Restart Supabase and apply migrations
npx supabase stop && npx supabase start
npx supabase migrations up

Then define your workflow (full guide):

import { Flow } from '@pgflow/dsl';

new Flow<{ url: string }>({ slug: 'analyzeArticle' })
  .step({ slug: 'scrape' }, (flowInput) => scrapeWebsite(flowInput.url))
  .step({ slug: 'summarize', dependsOn: ['scrape'] }, (deps) =>
    summarize(deps.scrape)
  )
  .step({ slug: 'extractKeywords', dependsOn: ['scrape'] }, (deps) =>
    extractKeywords(deps.scrape)
  )
  .step(
    { slug: 'publish', dependsOn: ['summarize', 'extractKeywords'] },
    (deps) =>
      publish({ summary: deps.summarize, keywords: deps.extractKeywords })
  );

This replaces ~240 lines of queue setup, state management, and coordination code. See full comparison

Building workflows in Supabase today means wiring together pgmq, pg_cron, state tables, and Edge Functions yourself. It works, but it's tedious.

pgflow gives you:

• Declarative workflows - Define steps and dependencies in TypeScript. pgflow handles queues, state, and coordination.
• Built for Supabase - Runs entirely in your existing project. No Redis, no Temporal, no external services.
• AI-ready - Automatic retries with exponential backoff for flaky LLM APIs. Per-step, not per-workflow.
• Parallel processing - Fan out over arrays with independent retries. If 3 of 100 items fail, only those 3 retry.
• Full observability - All workflow state in Postgres. Query runs, debug failures, inspect outputs with SQL.
• Flexible triggers - Start from your app, database triggers, pg_cron, or direct SQL calls.

• AI Pipelines - Scrape websites, chunk content, generate embeddings, summarize with LLMs. Each step retries independently when APIs flake.
• Background Jobs - Process uploads, send emails, sync data. Reliable task queue processing without Redis or external services.
• RAG Pipelines - Chunk documents, generate embeddings, index content. Perfect for AI applications with multi-step LLM chains.
• Data Workflows - ETL pipelines, scheduled imports, multi-step transformations. All orchestrated in Postgres.

See how pgflow compares to Trigger.dev, Inngest, DBOS, and Vercel Workflows.

1. Define workflows using the TypeScript DSL
2. Deploy as Supabase Edge Functions
3. Compile & verify automatically in the database at worker startup
4. Trigger from your app, SQL, or pg_cron

The execution engine handles scheduling, retries, and result aggregation automatically.

Package	Version	Description
pgflow		CLI for installing pgflow into your Supabase project
@pgflow/core		SQL Core - foundational tables and functions
@pgflow/dsl		TypeScript DSL for defining flows with type inference
@pgflow/edge-worker		Task queue worker for Supabase Edge Functions
@pgflow/client		TypeScript client for starting and monitoring workflows

• Release Process: See RELEASES.md for how versions are managed and published
• Snapshot Releases: See SNAPSHOT_RELEASES.md for testing changes before release