An AI Agent that grows with you. πͺ΄
AI should feel persistent, not purely transactional.
Startfire acts as a companion operating system rather than a standard chatbot. It builds a continuously evolving understanding of who you are, retaining context across days, weeks, and months. As it learns your capabilities and goals, it autonomously navigates opportunities, scouts for roles, and curates pathways tailored uniquely to you.
| Capability | Description |
|---|---|
| π§ Persistent Memory | Remembers conversations, goals, and context across extended timeframes. |
| π― AI Job Discovery | Autonomously searches, ranks, and retrieves global opportunities daily. |
| π Semantic Retrieval | Fetches perfectly tailored information using advanced vector search. |
| π€ Agent Orchestration | A "Boss AI" manages highly specialized, ephemeral worker agents. |
| π Resume Optimization | Dynamically crafts and optimizes applications for specific roles. |
| βοΈ Outreach Automation | Prepares intelligent, personalized founder and recruiter outreach. |
| π± Telegram Assistant | Always-on messaging interface for immediate access and updates. |
| π Multi-platform | Extensible architecture built for the future of ambient computing. |
Built for scale, resilience, and complex asynchronous execution.
System Layers:
- Web Layer: Next.js interface for user management and high-level dashboard visualization.
- Core Orchestration: The "Boss AI" handling routing, memory states, and decision logic.
- Worker Agents: Specialized node instances spawned for discrete tasks (e.g., job scraping, resume tailoring) that terminate upon completion.
- Queues: BullMQ combined with Redis to handle asynchronous agent dispatching.
- Data & Memory: PostgreSQL for relational structure, Qdrant for vector embeddings, and Redis for high-speed caching.
An intelligent, multi-tiered approach to user context retention.
Our proprietary memory flow involves:
- Short-term Memory: Immediate session context and recent conversational turns.
- Long-term Relational: Structured data (career history, explicit preferences).
- Semantic Memory: Vectorized traits, nuanced skills, and inferred goals.
- Retrieval Flow: The Boss AI synthesizes state from all three tiers before dispatching instructions to worker agents.
A clean lifecycle from user intent to execution.
1. Input: Telegram/Web β 2. Processing: Core "Boss" AI β 3. Dispatch: Queue β 4. Execution: Worker Agents β 5. Context Update: Memory Systems β 6. Output: Response/Action.
A robust relational state management underlying the entire orchestration network.
| Category | Technologies |
|---|---|
| π Frontend | Next.js, TailwindCSS, TypeScript |
| βοΈ Backend | Express.js, Node.js |
| π§ AI Layer | LangGraph, OpenAI APIs |
| βοΈ Infrastructure | PostgreSQL, Redis, BullMQ, Qdrant, Prisma, Docker |
| π Workflow | Turborepo, pnpm |
| π€ Automation | Playwright |
| π Monitoring | Sentry, Pino |
Startfire/
βββ apps/
β βββ web/ # Next.js frontend application
β βββ core/ # Express API & Boss AI orchestrator
β βββ telegram/ # Telegram bot integration layer
β βββ workers/ # Modular, task-specific worker agents
βββ packages/
β βββ agents/ # Shared LangGraph agent logic
β βββ ai/ # LLM wrappers and utility functions
β βββ browser/ # Playwright automation scripts
β βββ config/ # Global TypeScript/ESLint configurations
β βββ database/ # Prisma clients and repository patterns
β βββ memory/ # Qdrant vector store and retrieval logic
β βββ shared/ # Common types, schemas, and utilities
βββ infra/ # Docker compose, scripts, and monitoring
βββ prisma/ # PostgreSQL database schema and migrations
- Node.js (v20+)
- pnpm (v9+)
- Docker & Docker Compose
# Clone the repository
git clone https://github.com/StarDust130/Startfire.git
cd Startfire
# Install dependencies
pnpm install
# Start infrastructure (PostgreSQL, Redis, Qdrant)
docker compose up -d
# Sync database
pnpm db:push
# Launch development servers
pnpm dev⦠Environment Setup
Duplicate the example environment file and fill in your credentials.
cp .env.example .envEssential Variables:
# Database
DATABASE_URL="postgresql://user:password@localhost:5432/Startfire"
# Memory & Queues
REDIS_URL="redis://localhost:6379"
QDRANT_URL="http://localhost:6333"
# Intelligence
OPENAI_API_KEY="sk-..."β¦ Docker Infrastructure
The project uses Docker Compose to containerize local dependencies required by the orchestrator:
- PostgreSQL: Primary relational datastore (Prisma).
- Redis: Powers BullMQ task queues and short-term caching.
- Qdrant: High-performance vector database for semantic memory retrieval.
Run docker compose up -d in the root directory to spin up the required backing services before starting the node instances.
- Core orchestrator architecture
- Basic persistent memory implementation
- Telegram bot integration
- WhatsApp integration
- Advanced autonomous workflows
- Voice interface capabilities
- Deeply integrated AI productivity OS features
- Browser automation for complex job applications
- Memory evolution and continuous self-reflection
Startfire is not trying to become another generic chatbot.
It is designed to become a deeply personalized, reliable, and adaptive digital twin. It focuses on continuity, trust, and proactive execution, moving beyond reactive Q&A to true asymmetric leverage for your future.
"We are entering an era where your OS doesn't just run your appsβit runs your opportunities."
This project is licensed under the StarDust License - "use it as you want but don't be a jerk." ~ π
β¨ β¨ β¨
Built with precision. Engineered for the future.
