Presk：AI 原生生产力平台，集成日历、项目、文件与消息

AI‑Native Productivity Platform

Presk is an AI‑first workspace that automates your workflow, centralizes your tasks, and connects your team in one place. It combines calendars, projects, files, and messaging with built‑in AI assistants to help you move faster and stay aligned.

What you can do with Presk

Calendars – Schedule meetings, set reminders, and manage events across teams.
Projects – Create projects, assign tasks, track progress, and manage deadlines.
Files – Create, edit, and co‑author documents in real time with your team and AI.
Messages – Chat with teammates and AI assistants in shared channels and threads.
Dashboard – See all your tasks, projects, and team activity in a single, customizable view.

Why Presk

AI‑native: AI is embedded in every surface—drafting docs, summarizing threads, suggesting next actions, and automating repetitive work.
Unified workspace: No more switching between tools; everything you need lives in one platform.
Team‑first: Built for collaboration, with real‑time editing, shared context, and clear ownership.

Tech stack

Framework: Next.js (App Router) + React 19
Monorepo: Turborepo with npm workspaces
Database: PostgreSQL via Prisma 7 (driver adapter @prisma/adapter-pg)
Auth: Better Auth
UI: shadcn/ui + Base UI + Tailwind CSS
AI: OpenAI Agents SDK, exposed to clients over MCP
i18n: next-intl (English and French)

Repository structure

Path	Description
`apps/web`	Next.js application (App Router, i18n, API + MCP routes).
`packages/ui`	Shared UI component library.
`packages/eslint-config`	Shared ESLint configuration.
`packages/typescript-config`	Shared TypeScript configuration.

Getting started

Prerequisites

Node.js >= 20.9.0
npm 11 (declared via packageManager; run corepack enable to match)
Docker (recommended) for a local PostgreSQL instance

Setup

Install dependencies from the repository root:
```
npm install
```
Start PostgreSQL:
```
docker compose up -d postgres
```
Create your environment file and fill in the values:
```
cp apps/web/.env.example apps/web/.env
```
At minimum set DATABASE_URL and BETTER_AUTH_SECRET. An OPENAI_API_KEY is only required for AI assistant features. See Environment variables below.

Generate the Prisma client and apply migrations:

npm run generate --workspace=web
npx prisma migrate deploy --schema apps/web/prisma/schema.prisma

Start the dev server:
```
npm run dev
```
The app is available at http://localhost:3000.

Prefer containers? docker compose up --build runs PostgreSQL, applies migrations, and serves the production build on port 3000.

Environment variables

All variables live in apps/web/.env (see apps/web/.env.example).

Variable	Required	Description
`DATABASE_URL`	Yes	PostgreSQL connection string.
`DATABASE_POOL_MAX`	No	Max Postgres connections per instance (default `3`).
`BETTER_AUTH_SECRET`	Yes	Secret used to sign sessions.
`BETTER_AUTH_URL`	Yes	Public base URL of the app.
`OPENAI_API_KEY`	No	Enables AI assistant features.
`OPENAI_API_URL`	No	Custom OpenAI-compatible base URL.
`ASSISTANT_MODEL`	No	Model used by the assistant (default `gpt-4o-mini`).
`MCP_RESOURCE_URL`	No	Public identifier of the OAuth2-protected `/mcp` endpoint (defaults to `${BETTER_AUTH_URL}/mcp`).

Available scripts

Run these from the repository root; Turborepo fans them out to the relevant workspaces.

Command	Description
`npm run dev`	Start the dev servers in watch mode.
`npm run build`	Build all workspaces.
`npm run lint`	Lint all workspaces.
`npm run typecheck`	Type-check all workspaces.
`npm run format`	Format with Prettier.

File storage

Uploaded files are stored as PostgreSQL large objects (pg_largeobject) in the same database as their metadata. There is no external object storage (Vercel Blob has been removed).

Bytes → large objects. lib/large-object.ts streams bytes into a large object and streams them back out. Responses use Readable.toWeb(...); nothing is buffered in memory.
Chunked uploads (Vercel-safe). Vercel caps a Function request body at 4.5 MB, so a file is uploaded in 3 MiB chunks under an UploadSession instead of one request:
- POST /api/files/uploads — authz + quota check, creates the large object and an UploadSession (name, parentId, mimeType, expectedSize, transient).
- PUT /api/files/uploads/[uploadId] — appends one raw chunk; the x-chunk-offset header must equal the bytes already received (rejects gaps, duplicates, and reordering). Retries are safe.
- POST /api/files/uploads/[uploadId]/complete — re-reads the stored bytes to validate size + SHA‑256, then creates the FileNode/FileVersion (or returns the oid for a transient message attachment).
- DELETE /api/files/uploads/[uploadId] — aborts and unlinks the large object.
- GET /api/files/uploads/sweep — unlinks sessions older than their lease (CRON_SECRET-guarded when set; also runs opportunistically on init).
Metadata → Prisma. FileNode, FileVersion, and MessageAttachment hold oid BigInt? @unique, sizeBytes, mimeType, and sha256. UploadSession tracks in-progress uploads.
One shared pool. lib/prisma.ts exports a single pg.Pool handed to new PrismaPg(pool, { disposeExternalPool: false }); Prisma and the large-object code share it. Each large-object operation runs on one dedicated connection inside a single BEGIN … COMMIT transaction (never pool.query()).
Downloads stream the large object with Content-Length/Content-Disposition; maxDuration is set on the upload/download routes.

Operational caveats

Connections are held for the duration of a transfer. Each in-flight upload/download occupies one pooled connection (default pool max is 3). Raise DATABASE_POOL_MAX if transfers run concurrently with Prisma traffic, or use a separate pool for large objects. Point DATABASE_URL at a pooler in production.
No fixed per-file cap. Uploads are bounded only by the organization's remaining storage quota (free 1 GiB / team 100 GiB / enterprise 1 TiB). A large download holds a pooled connection for the whole transfer, so very large files are best served from a dedicated media host.
In-memory retry only. A failed chunk is retried within the session; uploads do not resume across page reloads. Abandoned sessions are unlinked after their lease expires.
No Blob migration. Existing storageKey values were dropped in the 20261009140000_replace_blob_with_large_object migration; old Blob bytes are not imported.
Deletes unlink the backing large object(s) synchronously on a best-effort basis and log failures without failing the request.
SSL warning. Set sslmode=verify-full (or uselibpqcompat=true&sslmode=require) in DATABASE_URL to silence the pg-connection-string deprecation warning on cold start.

Verifying

apps/web/scripts/verify-large-object.ts uploads ≥8 MiB of random bytes, reads them back, and asserts equal size and SHA‑256 for both the single-shot and the chunked create/append/hash paths; it then deletes/aborts and asserts pg_largeobject_metadata returns to its baseline (no orphans):

npm run verify:large-object --workspace=web

Contributing

Contributions are welcome! Please read CONTRIBUTING.md for setup instructions and our workflow, and note our Code of Conduct. To report a security issue, see SECURITY.md.

License

Released under the MIT License.