Skip to main content
ProBot is built in 7 incremental stages. Stage 1 is shipped; Stages 2–7 are planned. The detailed per-stage breakdown lives in claude/plan.md in the repo.

​
Stage 1 - shipped ✅

The current public surface.
  • Scaffolding: Next.js 14 App Router + TypeScript strict
  • Drizzle schema + migrations (users, bots)
  • NextAuth email/password (Credentials provider, JWT)
  • Login + register UI
  • Multi-provider LLM clients: Anthropic + OpenAI + Azure (Google stub)
  • ProviderError taxonomy (invalid_key / rate_limit / provider_unavailable / timeout)
  • BYO-key transport (x-llm-api-key header)
  • Browser key store (localStorage)
  • 5-step bot factory
  • POST /api/bots upsert
  • Chat UI with safe markdown
  • POST /api/chat/[botId] with full request lifecycle, input + output sanitization, 2-tier rate limit
  • 260 tests across 24 files

​
Stage 2 - planned

Knowledge ingestion. PDF and URL imports.
  • AWS S3 (Always Free tier) for upload storage via presigned URLs
  • pdf-parse for text extraction
  • cheerio for URL → HTML → text
  • tiktoken for chunking by token budget
  • Persisted as additional bot-context segments

​
Stage 3 - planned

RAG / vector search.
  • Pinecone OR pgvector (decision pending - pgvector likely for free-tier alignment)
  • Embedding step at ingest time
  • Vector retrieval at chat time, top-k injected into system prompt
  • Falls back to full-context mode when vector store unavailable

​
Stage 4 - planned

Public chat without auth.
  • /u/[username]/chat route (already scaffolded as placeholder)
  • Conversation logging (anonymized, owner-readable in dashboard)
  • Per-IP rate limit on top of per-bot

​
Stage 5 - planned

Embeddable <script> widget.
  • Built as a standalone bundle, served via AWS CloudFront (Always Free tier)
  • One-line embed: <script src="https://d…cloudfront.net/widget.js" data-bot="…"></script>
  • iframe-isolated chat UI; bot owner’s BYO key not exposed to the embedding site (visitor uses owner’s hosted chat endpoint instead)

​
Stage 6 - planned

Dashboard, analytics, lead capture.
  • Bot owner dashboard with conversation history, message counts, top questions
  • In-app notifications with unread badge when a recruiter leaves an email mid-chat
  • Bot toggle (is_active)
  • Multi-bot support (planned but not in Stage 1)

​
Stage 7 - planned

Production hardening.
  • OAuth providers: Google, GitHub, LinkedIn
  • Email verification + password reset
  • Landing page polish
  • Redis-backed rate limiting (Upstash)
  • GDPR data-export + account deletion
  • One-command Docker self-host
  • Sentry error reporting with PII scrubbing

​
Where to read more

​
How to help

PRs for any planned stage are welcome. Start with CONTRIBUTING.md and look for good first issue.