Launchstack - Professional Document Reader AI

Launchstack is a Next.js platform for role-based document management, AI-assisted Q&A, and predictive document analysis. It combines document upload, optional OCR, embeddings, and retrieval to help teams find gaps and act faster.

Core Features

Clerk-based Employer/Employee authentication with role-aware middleware.
Document upload pipeline with optional OCR for scanned PDFs.
PostgreSQL + pgvector semantic retrieval for RAG workflows.
AI chat and predictive document analysis over uploaded content.
Agent guardrails with PII filtering, grounding checks, and confidence gating.
Supervisor agent that validates outputs against domain-specific rubrics.
Marketing pipeline with content generation for Reddit, X, LinkedIn, and Bluesky.
Optional web-enriched analysis with Tavily.
Optional reliability/observability via Inngest and LangSmith.

Predictive Analysis — Supported Document Types

Launchstack runs domain-specific analysis tailored to your document type:

| Type | What It Detects | |------|----------------| | Contract | Missing exhibits, schedules, addendums, and supporting agreements | | Financial | Missing balance sheets, audit reports, income statements | | Technical | Missing specifications, manuals, diagrams, deliverables | | Compliance | Missing regulatory filings, certifications, policy documents | | Educational | Missing syllabi, handouts, readings, linked resources | | HR | Missing policies, forms, benefits materials, handbooks | | Research | Missing cited papers, datasets, supplementary materials | | General | Any document with cross-references and attachments |

Each analysis type also extracts insights (deadlines, action items, resources, caveats) and runs chain-of-verification on high-priority predictions.

Importing External Knowledge

Launchstack can ingest content exported from third-party tools. No API keys or OAuth setup required — export your data, upload the files, and the ingestion pipeline handles the rest.

Supported Export Formats

| Source | Export Method | Resulting Format | Launchstack Adapter | |--------|-------------|-----------------|----------------| | Notion | Settings > Export > Markdown & CSV | .md, .csv (ZIP) | TextAdapter, SpreadsheetAdapter | | Notion | Page > Export > HTML | .html | HtmlAdapter | | Google Docs | File > Download > Microsoft Word | .docx | DocxAdapter | | Google Sheets | File > Download > CSV or Excel | .csv, .xlsx | SpreadsheetAdapter | | Google Drive | Google Takeout (takeout.google.com) | .docx (ZIP) | DocxAdapter | | Slack | Workspace Settings > Import/Export > Export | .json (ZIP) | JsonExportAdapter | | GitHub | Code > Download ZIP | .md, .txt (ZIP) | TextAdapter | | GitHub | gh issue list --json ... | .json | JsonExportAdapter | | GitHub | gh pr list --json ... | .json | JsonExportAdapter |

How to Export

Notion

Open your Notion workspace.
Click the ... menu on a page, or go to Settings & members > Export for a full workspace export.
Select Markdown & CSV as the format and check Include subpages if needed.
Download the ZIP and upload it directly to Launchstack.

Google Docs / Sheets

Open the document in Google Docs or Sheets.
Go to File > Download and choose Microsoft Word (.docx) or CSV / Excel (.xlsx).
Upload the downloaded file. For bulk exports, use Google Takeout to export your Drive as a ZIP.

Slack

Go to Workspace Settings > Import/Export Data > Export.
Choose a date range and start the export.
Download the ZIP and upload it to Launchstack. Each channel's messages will be ingested as a separate document.

GitHub

Repo docs: Click Code > Download ZIP on any GitHub repository. Upload the ZIP — all Markdown and text files will be ingested.

Issues: Install the GitHub CLI and run:

gh issue list --state all --limit 1000 --json number,title,body,state,labels,author,createdAt,closedAt,comments > issues.json

Upload the resulting issues.json file.

Pull requests: Run:

gh pr list --state all --limit 1000 --json number,title,body,state,labels,author,createdAt,mergedAt,comments > prs.json

Upload the resulting prs.json file.

All uploaded content flows through the standard ingestion pipeline (chunking, embedding, RAG indexing) and becomes searchable alongside your other documents.

Architecture

Launchstack follows a three-layer modular architecture:

block-beta
  columns 9

  SLABEL["Services\nLayer"]:1
  MKT["Marketing Engine\n─────────────\nTrend Analysis\nContent Generation\nWeb Scraping Jobs"]:2
  LEG["Legal Services\n─────────────\nTemplate Library\nAuto-Fill & Clauses\nLegal Vault"]:2
  ONB["Employee Onboarding\n─────────────\nOnboarding Agent\nQuizzes & Checks\nProgress Tracking"]:2
  DOCR["Document Reasoning\n─────────────\nPage Index & TOC\nRLM Agent\nKnowledge Graph"]:2

  space:9

  TLABEL["Tools\nLayer"]:1
  RAG["RAG Pipeline\n(BM25 + Vector)"]:2
  WEB["Web Search\n(Tavily, Firecrawl)"]:2
  REW["Doc Rewrite\n(Summarize, Refine)"]:2
  TMPL["Template Engine\n(Form → PDF)"]:2
  space:1
  ING["Doc Ingestion\n(OCR, Chunk, Embed)"]:4
  ENT["Entity Extraction\n(NER, Graph RAG)"]:4

  space:9

  PLABEL["Physical\nLayer"]:1
  DB["PostgreSQL + pgvector\n─────────────\nEmbeddings Index\nDocument Structure\nKnowledge Graph\nDomain Tables"]:2
  HOST["Hosting & Compute\n─────────────\nNext.js 15\nInngest Jobs\nAgent Hosting\nML Sidecar"]:2
  EXT["External Services\n─────────────\nOCR Providers\nFile Storage (S3)\nClerk Auth + RBAC"]:2
  KBS["Knowledge Bases\n─────────────\nCompany KB\nLegal Templates\nOnboarding Docs"]:2

  %% Service → Tool edges
  MKT --> RAG
  MKT --> WEB
  MKT --> REW
  LEG --> RAG
  LEG --> REW
  LEG --> TMPL
  ONB --> RAG
  ONB --> REW
  DOCR --> RAG
  DOCR --> WEB
  DOCR --> REW
  DOCR --> ING
  DOCR --> ENT

  %% Tool → Physical edges
  RAG --> DB
  RAG --> KBS
  WEB --> HOST
  REW --> HOST
  TMPL --> EXT
  TMPL --> KBS
  ING --> DB
  ING --> EXT
  ING --> HOST
  ENT --> DB
  ENT --> HOST

  classDef layer fill:#1a1a2e,color:#eee,stroke:none
  classDef svc fill:#4A90D9,color:#fff,stroke:#2C5F8A,stroke-width:1px
  classDef tool fill:#F5A623,color:#fff,stroke:#C47D0E,stroke-width:1px
  classDef phys fill:#27AE60,color:#fff,stroke:#1E8449,stroke-width:1px

  class SLABEL,TLABEL,PLABEL layer
  class MKT,LEG,ONB,DOCR svc
  class RAG,WEB,REW,TMPL,ING,ENT tool
  class DB,HOST,EXT,KBS phys

The platform is organized into:

Services Layer - Vertical business modules (Marketing, Legal, Onboarding, Document Reasoning)
Tools Layer - Reusable AI capabilities (RAG, Web Search, Document Processing, Entity Extraction)
Physical Layer - Infrastructure (PostgreSQL + pgvector, Next.js hosting, External services, Knowledge bases)

All services operate within domain-partitioned boundaries enforced by Clerk RBAC. RAG queries are scoped by domain + company_id to ensure data isolation.

Tech Stack

Next.js 15 + TypeScript
PostgreSQL + Drizzle ORM + pgvector
Clerk authentication
OpenAI + LangChain
UploadThing + optional OCR providers
Tailwind CSS

Prerequisites

Node.js 18+
pnpm
Docker + Docker Compose (recommended for local DB/full stack)
Git

Quick Start

1) Clone and install

git clone <repository-url>
cd pdr_ai_v2-2
pnpm install

2) Configure environment

Create .env from .env.example and fill required values:

DATABASE_URL
NEXT_PUBLIC_CLERK_PUBLISHABLE_KEY
CLERK_SECRET_KEY
BLOB_READ_WRITE_TOKEN (Vercel Blob read/write token)
OPENAI_API_KEY
INNGEST_EVENT_KEY, as placeholder

Optional integrations:

NODE_ENV=development (for development, otherwise assumed to be production)
UPLOADTHING_TOKEN
TAVILY_API_KEY
INNGEST_EVENT_KEY, INNGEST_SIGNING_KEY
AZURE_DOC_INTELLIGENCE_ENDPOINT, AZURE_DOC_INTELLIGENCE_KEY
LANDING_AI_API_KEY, DATALAB_API_KEY
LANGCHAIN_TRACING_V2, LANGCHAIN_API_KEY, LANGCHAIN_PROJECT
DEBUG_PERF (1 or true) to enable dev perf logs for middleware and key auth/dashboard APIs
SIDECAR_URL
NEO4J_URI
NEO4J_USERNAME
NEO4J_PASSWORD

2.1) Configure Vercel Blob Storage

Vercel Blob is used for storing uploaded documents. Both public and private stores are supported -- the upload logic auto-detects which mode the store uses and adapts automatically.

In the Vercel dashboard, go to Storage → Blob → Create Store.
Choose either Public or Private access. Both work:
- Public stores produce URLs the browser can load directly (faster for previews).
- Private stores keep files behind authentication; the app proxies content through /api/documents/[id]/content and /api/files/[id] so previews still work.
Generate a Read/Write token for the store and add it as BLOB_READ_WRITE_TOKEN in your environment (.env locally, or Vercel Project Settings for deploys).
Redeploy so the token is available at build and runtime.
Verify: sign in to the Employer Upload page, upload a small PDF, and confirm /api/upload-local returns a vercel-storage.com URL without errors.

3) Start database and apply schema

pnpm db:push

4) Run app

pnpm run dev

Open http://localhost:3000.

Docker Deployment Methods

Method 1: Full stack (recommended)

Runs db + migrate + app via Compose:

docker compose --env-file .env --profile dev up

Detached mode:

docker compose --env-file .env --profile dev up -d

Method 2: App container only (external DB)

Use this when your database is managed externally.

docker build -t pdr-ai-app .
docker run --rm -p 3000:3000 \
  -e DATABASE_URL="$DATABASE_URL" \
  -e

LaunchStack

Install / Use

README