Fullsend

An autonomous GTM agent that ships ideas continuously, builds its own tools, and gets smarter over time.

The Fullsend dashboard showing Discord integration and live message flow between services

See VISION.md for the full concept and architecture.

---

The Agent That Builds Its Own Sales Pipeline

Here's something wild that happened: Fullsend autonomously designed and built an entire hiring-signal outreach pipeline — from scratch, without human intervention.

What the Agent Did

The Orchestrator received a strategic goal: "Find companies actively hiring for roles that suggest they need AI/automation help, and reach out to decision-makers."

Here's what happened next:

1. The Agent Identified a Gap

Fullsend realized it didn't have a tool to find companies with active job postings. Instead of asking for help, it dispatched a request to the Builder service with a detailed PRD.

2. Builder Created the Tool

The Builder agent (powered by Claude Code) autonomously wrote job_posting_finder.py — a complete tool that:

• Scrapes job boards for relevant postings
• Extracts company names, roles, and hiring signals
• Scores companies based on ICP fit (team size, funding stage, tech stack)
• Returns structured data ready for outreach

3. FULLSEND Designed the Experiment

With the new tool available, the FULLSEND agent designed a multi-phase pipeline:

phases:
  - phase: 1_discover_companies
    tool: job_posting_finder
    # Find companies hiring for ops/automation roles
    
  - phase: 2_find_contacts  
    tool: browserbase_email_finder
    # Get HR/hiring manager emails
    
  - phase: 3_personalized_outreach
    tool: cold_email_sender
    # Send tailored pitches based on their job postings

4. Scheduled for Autonomous Execution

The experiment was registered with a cron schedule: 0 9 * * MON — every Monday at 9 AM, the pipeline runs automatically, finding fresh hiring signals and reaching out to new prospects.

The Compound Effect

This is the Fullsend loop in action:

Idea → Build Tool → Design Experiment → Execute → Learn → Repeat

The agent doesn't just run campaigns — it builds the infrastructure to run campaigns. Each tool it creates becomes available for future experiments. Each experiment generates learnings that inform the next strategy.

No human wrote the job posting scraper. No human designed the outreach sequence. The agent identified the need, built the solution, and deployed it.

This is what autonomous GTM looks like.

---

Quick Start

# Install uv (if not already installed)
curl -LsSf https://astral.sh/uv/install.sh | sh

# Clone and setup
git clone <repo-url>
cd fullsend
uv sync

# Configure
cp .env.example .env
# Edit .env with your API keys (Discord, Anthropic, Google, etc.)

# Define your product
# Edit context/product_context.md with what you're selling

# Run everything
./run_all.sh

# Open dashboard
open http://127.0.0.1:8050/

Fresh Start / New Product

# Reset everything and start fresh
./restart.sh

# Or keep product context, reset agent state
./restart.sh --soft

Project Structure

fullsend/
├── services/
│   ├── discord/       # Discord bot + Web dashboard
│   ├── watcher/       # Message classifier (Gemini)
│   ├── orchestrator/  # Strategic brain (Claude)
│   ├── executor/      # Runs experiments
│   ├── fullsend/      # Experiment designer (Claude Code)
│   ├── builder/       # Tool builder (Claude Code)
│   └── redis_agent/   # Metrics monitor
├── tools/             # Agent-built tools live here
├── context/           # Product context, learnings, worklist
└── demo/dashboard/    # Real-time visualization

Services

Discord (services/discord/)

Communication interface - Discord bot + web dashboard that connects humans to the agent.

# Run Discord bot only
ENV=discord uv run python -m services.discord.main

# Run web dashboard only
ENV=web uv run python -m services.discord.main

# Run both
ENV=both uv run python -m services.discord.main

Features:

• Slash commands: /status, /pause, /go, /idea, /focus, /learn, /wtf
• Ambient listening in configured channels
• Emoji reactions on captured ideas
• Action requests with human-in-the-loop
• Real-time web dashboard at http://localhost:8000

Development

# Install all dependencies
uv sync

# Add a dependency to a service
uv add <package> --package fullsend-discord

# Run tests
uv run pytest

# Lint
uv run ruff check .

Requirements

• Python 3.11+
• Redis (for pub/sub between services)
• Discord bot token (create at discord.com/developers)