Read It For Me

Personal daily digest for links, pasted text, and PDFs. Each item answers what changed, why it matters, what to do, and whether it is worth reading fully.

Live demo

Built with Svelte 5, Express, Render Workflows, and Render Postgres.

Inference by Together AI.

Table of contents

• Highlights
• Overview
• Architecture
• Usage
• Deploy on Render
• Configuration
• Operations
• Project structure
• Troubleshooting
• License

Highlights

• Action-oriented cards, not generic summaries: every item gets whatChanged, whyCare, whatToDo, and a read / skim / skip verdict.
• Live progress over Server-Sent Events while workflow tasks run on Render.
• Change tracking via Postgres snapshots: repeat sources compare against the last digest.
• Separated layers: Svelte UI, Express API, and workflow tasks live in distinct folders with shared types in shared/.
• Together AI stays in workflows: the web service orchestrates tasks; it never calls the LLM directly.

Overview

Read It For Me is for people who skim newsletters, release notes, and long threads and want a structured digest instead of a paragraph of fluff.

You submit URLs (one per line), pasted text blocks, optional PDFs, and an optional focus string (for example "AI infra" or "hiring"). The Express server starts Render Workflow tasks, polls until each completes, and streams status updates and per-item cards back to the browser. When all items are analyzed, a synthesis task produces a headline plus doToday, readNow, and skip lists.

The workflow service uses Together AI (default: meta-llama/Llama-3.3-70B-Instruct-Turbo) inside analyze_item and synthesize_digest. Prior snapshots from Postgres feed into the analyze prompt so the model can comment on what changed since the last run.

Current MVP gaps: PDF uploads are accepted but text extraction is a placeholder. The workflow service is created manually in the Render Dashboard (not in render.yaml), matching the pattern used in other Render Workflows examples in this workspace.

Architecture

Layer	Folder	Role
UI	frontend/	Svelte 5 form, SSE client, digest cards
API	server/	Express routes, orchestrator, Postgres, static SPA
Tasks	workflow/	fetch_item, analyze_item, synthesize_digest
Contracts	shared/	Types and Render URL helpers

Usage

Web UI

1. Open the deployed app.
2. Paste URLs, text, optional PDFs, and an optional focus.
3. Click Build digest. Cards appear as each item finishes; the summary panel loads at the end.

API (SSE)

POST /api/digest accepts multipart/form-data:

Field	Type	Notes
urls	string	Newline-separated URLs (max 20 lines)
text	string	Newline-separated text blocks
focus	string	Optional theme, max 500 chars
pdfs	file[]	Optional PDFs, max 5 files, 3 MB each

Response is text/event-stream with events: status, card, done, error.

Requires RENDER_API_KEY and DATABASE_URL on the web service. Returns 503 if either is missing.

Health

curl https://read-it-for-me.onrender.com/health
# {"ok":true}

Deploy on Render

Primary path: Blueprint for the web service and database, then a manual Workflow service.

1. Push to GitHub

Connect this repository to GitHub if it is not already.

2. Deploy the Blueprint

Use the Deploy to Render button in the app UI, or apply render.yaml from the Render Dashboard. This creates:

• Web service read-it-for-me — node dist/server/index.js, health check /health
• Postgres read-it-for-me-db — connection string wired to DATABASE_URL

Set RENDER_API_KEY on the web service (Account Settings → API Keys).

3. Create the Workflow service

The Blueprint does not include the workflow runner. In the Dashboard:

1. New → Workflow (link the same repo).
2. Build command: npm install && npm run build
3. Start command: node dist/workflow/index.js
4. Environment:
   • TOGETHER_API_KEY — required
   • DATABASE_URL — same Postgres as the web service (internal URL on the private network)
   • TOGETHER_MODEL — optional, defaults to Llama 3.3 70B Instruct Turbo
5. Note the workflow slug (default expectation: read-it-for-me-workflow). Set WORKFLOW_SLUG on the web service if yours differs.

4. Verify

• GET /health returns {"ok":true}
• Submit a digest from the UI with at least one URL
• Check workflow task logs in the Render Dashboard if cards never appear

Configuration

Web service (read-it-for-me)

Variable	Required	Default	If missing
RENDER_API_KEY	Yes (for digest)	—	/api/digest returns 503; /health still works
DATABASE_URL	Yes (for digest)	—	/api/digest returns 503; persistence disabled
WORKFLOW_SLUG	No	read-it-for-me-workflow	Task paths won't match your workflow
GITHUB_REPO_URL	No	https://github.com/ojusave/read-it-for-me	Deploy button points at default repo
POLL_INTERVAL_MS	No	3000	Task poll interval
PORT	No	3000	Set by Render in production
NODE_ENV	No	—	production enables Postgres SSL

Workflow service

Variable	Required	Default	If missing
TOGETHER_API_KEY	Yes	—	analyze_item / synthesize_digest fail
DATABASE_URL	Recommended	—	Snapshots not written from workflow side (web service handles persistence)
TOGETHER_MODEL	No	meta-llama/Llama-3.3-70B-Instruct-Turbo	Falls back to default model

Operations

• Health check: GET /health — used by Render on the web service.
• Logs: Render Dashboard → service → Logs. Workflow task output appears on the Workflow service.
• Database: Tables digest_runs, source_snapshots, digest_items are created on web service startup when DATABASE_URL is set.
• Ephemeral disk: Uploaded PDFs are read in memory and deleted; do not rely on local filesystem persistence on Render.
• Free tier: Web services spin down after inactivity; first request after idle may be slow. Free Postgres expires after 30 days.

Project structure

read-it-for-me/
├── frontend/src/          # Svelte UI (components, api client)
├── server/
│   ├── index.ts           # Express entry, static files, config route
│   ├── routes/            # health, digest
│   └── lib/               # db, orchestrator, sse
├── workflow/
│   ├── index.ts           # Task registration
│   ├── tasks/             # fetch, analyze, synthesize
│   └── lib/together.ts   # Together AI adapter
├── shared/                # types.ts, renderUrls.ts
├── static/images/         # README architecture diagrams
├── render.yaml            # Web + Postgres Blueprint
└── vite.config.ts         # Builds frontend → dist/client

Where to change things

• Digest prompts and LLM JSON shape: workflow/tasks/analyze.ts, workflow/tasks/synthesize.ts
• SSE event shape: server/lib/orchestrator.ts, shared/types.ts
• UI layout and streaming client: frontend/src/App.svelte, frontend/src/lib/api.ts
• Postgres schema: server/lib/db.ts

Troubleshooting

RENDER_API_KEY is not configured

Set the key on the web service. The server starts without it, but digest requests are blocked.

DATABASE_URL is not configured

Wire Postgres from the Blueprint on the web service. Digest requires persistence for snapshot diffs.

Tasks fail immediately / wrong workflow

Confirm WORKFLOW_SLUG on the web service matches the slug shown on your Workflow service in the Dashboard. Task names are {slug}/fetch_item, {slug}/analyze_item, {slug}/synthesize_digest.

Together errors

Check TOGETHER_API_KEY on the Workflow service and that the model name in TOGETHER_MODEL is available on your Together account.

PDF content is generic

Expected in MVP: fetch_item returns placeholder text for PDFs until real extraction is added.

Stale UI after deploy

Run npm run build before deploy. The web service serves dist/client; Vite dev server is not used in production.

License

This project is licensed under the MIT License.