Read It For Me

Paste links, notes, or PDFs. Get action-first digest cards and a short summary of what to do, what to read, and what to skip. Inference runs on Together AI inside Render Workflows.

What This Demo Shows

This repo demonstrates how to run multi-step chat inference on Render:

Platform	Role
Render Workflows	Durable fetch_item, analyze_item, and synthesize_digest tasks with retries and timeouts
Together AI	Chat completions for per-item cards and the digest summary; model chosen in the UI
Render Postgres	Digest runs, item results, and source snapshots
Render Web Services	Svelte UI, Express API, SSE streaming, and GET /api/models

Product tour

Compose a digest, watch the Workflow timeline, then read action-first cards. Screenshots from the live demo.

Compose

Paste sources, search Together chat models, then build a digest.

Results

The live Gantt shows fetch → analyze → synthesize. Cards lead with Do this, then why it matters, what changed, and a read / skim / skip call. The summary lists what to do today, read next, and skip.

How a digest runs

1. Browser starts a digest from the composer
2. Web service persists the run, streams SSE, and dispatches Workflow tasks
3. Render Workflows executes the pipeline:

Workflow task	What it does
fetch_item	Loads URL / text / PDF content for one source
analyze_item	Together chat → one action-first card
synthesize_digest	Together chat → headline plus doToday, readNow, skip

4. Cards and stage events stream into the UI as each task finishes

Quick Start

Prerequisites

• Render account
• Together AI API key
• Render API key

Deploy

1. Click Deploy to Render above (or apply render.yaml)
2. Set secrets when prompted: RENDER_API_KEY, TOGETHER_API_KEY
3. Create the Workflow service manually (Blueprints do not create Workflows yet):
   • Dashboard → New → Workflow (same repo, same region as web + Postgres)
   • Build: npm install && npm run build
   • Start: node dist/workflow/index.js
   • Slug: read-it-for-me-workflow (or set WORKFLOW_SLUG on the web service to match)
   • Env: TOGETHER_API_KEY, DATABASE_URL, optional TOGETHER_MODEL
4. Open the web service URL, paste a URL, and click Build digest

Use the same Together key on web (model catalog) and Workflow (inference). Confirm GET /api/models returns "source":"together".

Default model: meta-llama/Llama-3.3-70B-Instruct-Turbo. PDF uploads are accepted; text extraction is still an MVP placeholder.

Features

Feature	Description
Action-first cards	Each item leads with Do this, then why, what changed, and a read / skim / skip call
Together model picker	Searchable chat models from Together; selected model runs analyze and synthesize
Live Workflow timeline	SSE-driven left-to-right Gantt (fetch → analyze → synthesize)
Change tracking	Postgres snapshots compare repeat sources against the last digest
Split responsibilities	Web orchestrates; Together inference stays inside Workflow tasks

Configuration

Variable	Where	Description
DATABASE_URL	Web	Render Postgres connection string
TOGETHER_API_KEY	Web + Workflow	Web: model catalog. Workflow: inference
RENDER_API_KEY	Web	Start and poll workflow tasks
TOGETHER_MODEL	Both (optional)	Default chat model
WORKFLOW_SLUG	Web	Must match the Dashboard workflow slug
ENABLE_WORKFLOW_TIMELINE	Web	Set 0 to hide the Gantt

Learn More

Render: Workflows · Postgres · Deploy button · Discord

Together AI: Docs · Models

This repo: Live demo · GitHub

License

MIT