Prodify

An AI-powered product management workspace. Write PRDs, plan roadmaps, prioritise features and synthesise research in one place, with an AI co-pilot (Google Gemini) that streams its answer and writes it directly into your documents.

---

Features

AI workflows (system prompts in lib/prompts.ts)

• PRD — problem statement, success metrics, user stories, scope, requirements, open questions
• User Stories — Gherkin acceptance criteria, story points, dependencies, definition of done
• Roadmap — Now / Next / Later plan with sequencing rationale and dependencies
• Prioritization — RICE scoring, MoSCoW classification and trade-off analysis
• Research — turns interview notes and feedback into themes, quotes and opportunities
• Data — funnel analysis, anomalies, falsifiable hypotheses and experiment design
• Free Chat — general PM co-pilot with no fixed output format

Editor

• Rich text editing with Tiptap, auto-saved 2 seconds after you stop typing
• PRD, Stories, Roadmap and Research answers are written straight into a document
• Diff toggle on each AI answer shows which lines were added or removed
• Copy or download a document as Markdown or plain text

Workspace

• Folder tree in the left sidebar: create, rename and delete folders and documents
• Roadmap board, RICE / MoSCoW prioritisation panel, research insights and data panel
• Everything is saved to PostgreSQL through small REST routes

Tasks and GitHub

• Tasks tab: each task gets a key like PROD-42 and a "copy branch name" button (prod-42-add-csv-import)
• Connect a GitHub App: pull requests whose branch, title or description mention a task key are linked automatically, or paste a PR URL on a task
• PR badges on tasks (open / draft / merged / closed), a PR list in the task detail, and a Delivery strip on each PRD ("6 of 9 tasks shipped")
• Optional automation (off by default): PR opened → task to Review, PR merged → Done, always noted in the task's activity

Keyboard shortcuts

Shortcut	Action
Cmd+K	Command palette
Cmd+N	New note in the current folder
Cmd+Enter	Send AI message
Cmd+/	Cycle right panel tabs
?	Show all shortcuts

---

Tech Stack

Layer	Technology
Framework	Next.js 14 (App Router)
Language	TypeScript
Styling	Tailwind CSS, shadcn/ui
Rich text	Tiptap
State	Zustand
Database	PostgreSQL (Neon) via Prisma 5
AI	Google Gemini via @google/genai, streamed
Layout	react-resizable-panels

---

Getting Started

1. Clone and install

git clone https://github.com/Ratangulati/Prodify.git
cd Prodify
npm install

2. Environment variables

cp .env.example .env

Variable	Required	Description
GEMINI_API_KEY	Yes	Gemini API key from Google AI Studio
DATABASE_URL	Yes	Postgres connection string used by the app (Neon pooled URL)
DIRECT_URL	Yes	Postgres connection string used by Prisma migrations (Neon direct URL)
GEMINI_MODELS	No	Comma-separated models to try in order. Default: gemini-3.8-flash,gemini-3.6-flash,gemini-3.5-flash-lite
GITHUB_APP_ID, GITHUB_APP_SLUG, GITHUB_PRIVATE_KEY, GITHUB_WEBHOOK_SECRET	No	Only for pull-request linking. See GitHub pull-request linking

Use .env (not .env.local): the Prisma CLI only reads .env. Never commit it — it is in .gitignore.

3. Set up the database

npm run db:migrate   # prisma migrate deploy: creates the tables
npm run db:seed      # optional: demo documents, features and insights (run once, on an empty DB)

4. Run it

npm run dev

Open http://localhost:3000.

5. Tests

npm test   # Vitest: task keys, GitHub signatures, key matching, webhooks, manual linking

GitHub is never called in tests: requests go through a fake fetch, and the webhook tests use an in-memory stand-in for the database.

---

How the AI works

AIChatPanel ──POST /api/ai──▶ route.ts ──▶ lib/ai.ts ──▶ Gemini
     ▲                          │  (system prompt from lib/prompts.ts)
     └──── streamed text ◀──────┘

1. You pick a workflow tab (PRD, Stories, …) and type a prompt.
2. AIChatPanel sends the prompt, the conversation history for that workflow and (for document workflows) the current document text to /api/ai.
3. app/api/ai/route.ts builds the system prompt for the workflow and calls streamChat in lib/ai.ts.
4. lib/ai.ts tries each Gemini model in order. A 429 / 500 / 503 gets one retry after 1 second, then the next model is tried. A bad key or bad request stops straight away.
5. The text is streamed back to the browser with a ReadableStream and shown as it arrives.
6. If anything fails, the route returns a JSON error (or an error marker mid-stream) and the chat shows a clear message instead of a broken answer.
7. For PRD, Stories, Roadmap and Research, the finished Markdown is converted to HTML and written into the matching document (a new one is created if the open document is a different type). Prioritization, Data and Free Chat answer in the chat only.

Diff view

After an answer is written into a document, the chat shows a card with a Diff toggle. It compares the document before and after line by line using a longest-common-subsequence (LCS) diff, and colours added lines green and removed lines red. Documents over 400 lines skip the LCS step and show everything as removed then added, to keep it fast.

---

GitHub pull-request linking

Tasks show the pull requests that deliver them, and each PRD shows how much has shipped. Prodify uses a GitHub App: it only sees the repositories you pick, webhooks are built in, and no long-lived token is stored. The server signs a short-lived token with the app's private key and exchanges it for a one-hour installation token whenever it needs to call GitHub.

How PRs get linked

• Automatically: a PR whose branch name, title or description contains a task key (prod-42-add-csv-import, PROD-42: add CSV import, "Closes PROD-42") is linked to that task when GitHub sends the webhook.
• Manually: open a task and paste the PR's URL. Prodify checks with GitHub that the PR exists in a tracked repository.

A task counts as shipped when it's Done or one of its PRs is merged.

1. Create the GitHub App

GitHub → Settings → Developer settings → GitHub Apps → New GitHub App:

Setting	Value
Homepage URL	Your app URL, e.g. http://localhost:3000
Setup URL	<app URL>/api/github/setup and tick Redirect on update
Webhook URL	<public URL>/api/github/webhook (for local use, the smee.io URL from step 3)
Webhook secret	A long random string: openssl rand -hex 32
Repository permissions	Pull requests: Read-only (Metadata: Read-only is added automatically)
Subscribe to events	Pull request (installation events are always sent)
Where can it be installed	Only on this account (or any account, for an organisation)

After creating it, note the App ID and the app's URL name (the slug: github.com/apps/<slug>), then Generate a private key (a .pem file).

2. Environment variables

Print the private key as one line, with \n for each line break:

awk 'NF {printf "%s\\n", $0}' prodify-yourname.private-key.pem

Then add to .env (paste that output inside the quotes):

GITHUB_APP_ID=123456
GITHUB_APP_SLUG=prodify-yourname
GITHUB_WEBHOOK_SECRET=the-secret-from-step-1
GITHUB_PRIVATE_KEY="-----BEGIN RSA PRIVATE KEY-----\nMIIE...\n-----END RSA PRIVATE KEY-----\n"

Keep the key out of git (.env is ignored) and delete the downloaded .pem once it's in .env. Then apply the new tables with npm run db:migrate and restart the app.

3. Receive webhooks locally

GitHub can't reach localhost, so forward webhooks with smee.io:

# 1. Open https://smee.io/new and copy the channel URL into the app's Webhook URL
# 2. Forward it to the dev server:
npx smee-client --url https://smee.io/<your-channel> --target http://localhost:3000/api/github/webhook

(ngrok works too: ngrok http 3000, then use https://<id>.ngrok-free.app/api/github/webhook as the Webhook URL.) To test duplicate handling, open the app's Advanced tab on GitHub and click Redeliver on any delivery: the second copy is answered with {"status":"duplicate"}.

4. Connect

In Prodify, click the GitHub icon in the left sidebar → Connect GitHub, install the app on your repositories, and back in Prodify pick which repositories to track. Optionally turn on Move tasks automatically.

How the webhook stays safe

• Every delivery's X-Hub-Signature-256 is checked against the raw body with the webhook secret (constant-time compare); anything else gets a 401.
• Each delivery id is stored in the same transaction as its changes: a repeated delivery is skipped, and a failed one rolls back so redelivering it (the app's Advanced tab) works. GitHub doesn't retry failed deliveries on its own. Events older than the stored PR data are ignored.
• A repository can be tracked by only one workspace, and the repository decides which workspace's tasks a webhook may touch.

Limits worth knowing

• Prodify has no sign-in yet, so anyone who can open the app can change the GitHub settings. Only run it where that's acceptable.
• WebhookDelivery keeps every delivery id. Clear old ones occasionally: DELETE FROM "WebhookDelivery" WHERE "receivedAt" < now() - interval '30 days';
• GitHub API errors, including rate limits, are shown in the UI with when to retry.

---

Project structure

app/
  api/ai/route.ts        Gemini streaming endpoint
  api/db/*               REST routes per table (documents, features, filenodes, insights, messages)
  api/tasks/*            Tasks, and linking/unlinking PRs on a task
  api/github/*           Install + setup flow, repos, settings, webhook
  page.tsx, layout.tsx
components/
  AppShell.tsx           Three-panel layout + global shortcuts
  editor/                EditorPanel (title, export) and Editor (Tiptap)
  sidebar/               FileExplorer, AIChatPanel (workflows, diff), Research and Data panels
  tasks/                 Tasks board, task detail, PR badge
  roadmap/, prioritization/, modals/, ui/
lib/
  ai.ts                  Gemini client, model fallback, error messages (server only)
  ai-client.ts           Browser-side AI error helpers
  prompts.ts             System prompt per workflow
  store.ts               Zustand store; updates the UI first, then syncs to the API
  prisma.ts              Prisma client singleton
  tasks.ts, task-dto.ts  Task keys, branch names, "shipped" rule
  use-tasks.ts           Browser store for tasks (polls while visible)
  github/                App auth + API client, signature check, key matching, webhook, manual linking
  types.ts
prisma/
  schema.prisma, seed.ts, migrations/

---

Deploying to Vercel

1. Import the repo in Vercel (framework: Next.js).
2. Add GEMINI_API_KEY, DATABASE_URL and DIRECT_URL in Project Settings → Environment Variables (plus the four GITHUB_* variables if you use pull-request linking, with the app's Setup and Webhook URLs pointing at your Vercel domain).
3. Run npm run db:migrate once against the production database.

The build runs prisma generate && next build.