Skip to main content
Get Dream Foundry up and running in minutes. This guide walks you through executing your first competitive AI agent forge from scratch.

​
What You’ll Build

You’ll run a complete Dream Foundry workflow where 5 AI agents compete to generate a weekly “AI Events in the Bay Area” Discord post. The agents use different strategies (speed vs. quality vs. insider knowledge), and the best implementation wins.
The Demo Scenario: Multiple agents scrape AI/ML events from lu.ma and Meetup.com, get scored on quality, speed, and reliability, and the winner’s output gets published to Discord.

​
Prerequisites

Before you begin, ensure you have:
  • Python 3.12+ installed
  • Basic familiarity with command line
  • (Optional) API keys for advanced features:
    • DAYTONA_API_KEY - For isolated sandbox execution
    • SENTRY_DSN - For error monitoring
    • DISCORD_WEBHOOK_URL - For publishing results
    • ELEVENLABS_API_KEY - For winner announcement narration

​
Installation

1

Clone and navigate to the project

git clone <repository-url>
cd dream-foundry
2

Install dependencies

pip install -r requirements.txt
This installs core dependencies:
  • streamlit - Demo UI
  • beautifulsoup4, selenium, requests - Web scraping
  • sentry-sdk - Error monitoring
  • python-dotenv - Environment management
3

Configure environment (optional)

For the basic demo, no API keys are required. For full features:
cp .env.example .env
# Edit .env with your API keys
Without API keys, the forge runs in local mode (no Daytona sandboxes, no Discord publishing). This is perfect for testing!

​
Your First Forge Run

Run the forge with the default objective: 
python forge.py

python forge.py
# Uses: "Generate weekly 'AI Events in San Francisco' post for Discord"

​
What Happens During Execution

You’ll see a 5-phase workflow in your terminal:
1

Initialization

============================================================
DREAM FOUNDRY - Forge Loop
============================================================
Objective: Generate weekly 'AI Events in San Francisco' post for Discord
Candidates: 5
Execution: LOCAL
2

Running Candidates

Each of the 5 agents executes:
PHASE: Running Candidates
----------------------------------------

[Agent Alpha]
  [Run] Starting Agent Alpha...
  [Done] Output produced
  [Done] 4.2s

[Agent Beta]
  [Run] Starting Agent Beta...
  [Done] Output produced
  [Done] 28.5s

[Agent Gamma]
  [Gamma] Fetching from lu.ma and Meetup...
  [Gamma] Verified: Daytona HackSprint SF
  [Gamma] Verified: AI Makerspace Weekly Meetup
  [Gamma] Found 10 verified events
  [Done] 3.1s

[Agent Delta]
  [Error] Timeout after 60 seconds
  [Sentry] Error captured

[Agent Epsilon]
  [Done] Output produced
  [Done] 2.8s
3

Scoring

============================================================
PHASE: Scoring
----------------------------------------
Candidate     Success  Quality    Speed    TOTAL
----------------------------------------------
alpha        PASS     75.6       86.0     77.2
beta         PASS     91.5       5.0      61.9
gamma        PASS     94.4       89.7     94.4
delta        FAIL     0.0        0.0      0.0
epsilon      PASS     54.5       90.7     63.1

  Quality: 10/10 valid events, Hackathon: Yes
Scoring Breakdown:
  • Success (20%): Did it produce output without errors?
  • Quality (60%): Valid events, correct dates (Jan 24-31, 2026), hackathon included, proper URLs
  • Speed (20%): Faster execution = higher score
4

Winner Selection

============================================================
WINNER: gamma (score: 94.4)
============================================================

Scores written to: artifacts/scores.json
Winner written to: artifacts/winner.txt

Artifacts produced:
  - artifacts/alpha/discord_post.md [exists]
  - artifacts/beta/discord_post.md [exists]
  - artifacts/gamma/discord_post.md [exists]
  - artifacts/delta/discord_post.md [missing]
  - artifacts/epsilon/discord_post.md [exists]
  - artifacts/scores.json
  - artifacts/winner.txt

​
Inspect the Results

View the winning output: 
cat artifacts/gamma/discord_post.md

Example Output

# AI Events in the Bay Area
## Week of January 24-31, 2026

**Daytona HackSprint SF**
- Date: Saturday, January 24, 2026
- Time: 9:00 AM - 6:00 PM
- Location: San Francisco, CA
- Type: hackathon
- [RSVP](https://lu.ma/kga3qtfc) (via lu.ma)

**AI Makerspace Weekly Meetup**
- Date: Wednesday, January 28, 2026
- Time: 6:00 PM - 9:00 PM
- Location: San Francisco, CA
- Type: meetup
- [RSVP](https://lu.ma/aimakerspace) (via lu.ma)

...

---
*Found 10 verified events | Generated by Agent Gamma (The Insider)*

​
Option 2: Streamlit UI (Interactive Demo)

For a visual, 9-phase interactive demo: 
streamlit run app.py
This launches a browser-based UI at http://localhost:8501 with:
  1. The Dreamcatcher - Input your objective
  2. The Dream Factory - View candidate strategies
  3. The Arena - Watch agents compete in real-time
  4. Arena Results - See the scoreboard
  5. Forge Runoff - Polish - CodeRabbit improvements (simulated)
  6. Forge Runoff - Retest - Re-run with polished code
  7. The Podium - Final winner announcement
  8. Showcase - ElevenLabs narration + Discord publishing
The Streamlit UI includes simulated CodeRabbit and ElevenLabs features for demonstration purposes. The CLI mode (forge.py) runs the actual competitive scoring system.

​
Understanding the Candidates

Each agent uses a different strategy:

Agent Alpha

The Speed DemonFast HTTP scraping with BeautifulSoup. Can’t handle JavaScript-rendered content.
  • ⚡ ~4-5s execution
  • ❌ Misses JS pages

Agent Beta

The PerfectionistUses Selenium for full page loads. Sees everything a human would.
  • ✅ Handles JS
  • 🐌 ~25-30s execution

Agent Gamma

The Insider (Usually Wins)Uses verified event database + direct API calls. Fast and reliable.
  • ⚡ ~3s execution
  • ✅ 10/10 quality

Agent Delta

The CrasherIntentionally unstable to demonstrate Sentry error monitoring.
  • ❌ Crashes frequently
  • 📊 Errors captured by Sentry

Agent Epsilon

The HallucinatorGenerates plausible but unverified data. Fast but unreliable.
  • ⚡ Very fast
  • ❌ 0/5 valid events

​
Next Steps

Full Installation

Set up all API keys for Daytona sandboxes, Sentry monitoring, and Discord publishing

Architecture

Learn how the 5-phase Dream Foundry pipeline works

Scoring System

Understand how candidates are scored on success, quality, and speed

Custom Agents

Build your own competing agent implementations

​
Troubleshooting

Make sure you installed dependencies:
pip install -r requirements.txt
This usually means a network issue or rate limiting. Check:
  • Internet connection
  • Firewall settings
  • Rate limits on lu.ma/Meetup (wait 5 minutes and retry)
Sentry batches events asynchronously. The forge includes sentry_sdk.flush() to force immediate sending. Verify your SENTRY_DSN in .env.
Ensure:
  • DAYTONA_API_KEY is set in .env
  • You have an active Daytona account
  • Network allows connections to Daytona API
Fallback: Remove --daytona flag to run locally
Rate Limiting: The agents scrape public lu.ma and Meetup pages. If you run the forge repeatedly (>5 times in a minute), you may hit rate limits. Wait 5 minutes between runs or use the --daytona flag to isolate requests.

Build docs developers (and LLMs) love

Get started for freeTalk to us