Skip to main content

​
Overview

SkyDiscover includes a built-in live monitoring dashboard that provides real-time visualization of the evolutionary search process. The dashboard displays fitness progression, solution quality, and AI-generated summaries.

​
Basic Configuration

monitor:
  enabled: false
  port: 8765
  host: "127.0.0.1"
  max_solution_length: 10000

​
Enabling the Monitor

​
In Configuration File

configs/human_in_the_loop.yaml
monitor:
  enabled: true
  port: 8765
  host: "127.0.0.1"

​
From CLI

skydiscover-run program.py evaluator.py \
  -c configs/adaevolve.yaml \
  --monitor

​
Access the Dashboard

Once enabled, open your browser to: 
http://127.0.0.1:8765

​
MonitorConfig Parameters

Defined in skydiscover/config.py:500-515
​
enabled
bool
default:"false"
Enable or disable the live monitoring dashboard
​
port
int
default:"8765"
Port number for the web server
​
host
str
default:"127.0.0.1"
Host address to bind the server to. Use 0.0.0.0 to allow external access
​
max_solution_length
int
default:"10000"
Maximum character length to display for solutions in the dashboard

​
AI Summary Configuration

The monitor can generate AI-powered summaries of search progress using an LLM.
​
summary_model
str
default:"gpt-5-mini"
Model to use for generating progress summaries
​
summary_api_key
str
default:"None"
API key for summary model. Falls back to OPENAI_API_KEY if not specified
​
summary_api_base
str
default:"https://api.openai.com/v1"
API base URL for summary model
​
summary_top_k
int
default:"3"
Number of top programs to include in the summary
​
summary_interval
int
default:"0"
Auto-generate summary every N programs. Set to 0 for manual triggering only

​
Complete Example

configs/monitor_full.yaml
monitor:
  # Server settings
  enabled: true
  port: 8765
  host: "0.0.0.0"  # Allow external access
  max_solution_length: 15000
  
  # AI summary settings
  summary_model: "gpt-4o-mini"
  summary_api_key: ${OPENAI_API_KEY}
  summary_api_base: "https://api.openai.com/v1"
  summary_top_k: 5
  summary_interval: 10  # Generate summary every 10 programs

​
Dashboard Features

The live monitor provides:
  • Fitness progression over iterations
  • Best, median, and worst fitness curves
  • Island-specific plots (for multi-island algorithms)
  • Interactive zooming and panning
  • View all generated solutions
  • Syntax-highlighted code display
  • Fitness metrics for each solution
  • Filtering and sorting capabilities
  • High-level progress overview
  • Key insights from top solutions
  • Strategy recommendations
  • Automatic or manual generation
  • Total iterations completed
  • Best fitness achieved
  • Improvement rate
  • Time per iteration
  • Model usage breakdown

​
Human-in-the-Loop Integration

Combine the monitor with human feedback for interactive search:
configs/human_in_the_loop.yaml
# Enable both monitor and human feedback
monitor:
  enabled: true
  port: 8765

human_feedback_enabled: true
human_feedback_file: "feedback.txt"
human_feedback_mode: "append"

​
Providing Feedback

  1. View solutions in the dashboard
  2. Write feedback to the feedback file:
    feedback.txt
    The current solutions are too complex. Focus on simpler approaches.
Try using dynamic programming instead of recursion.
  3. Feedback is incorporated into the next generation

​
Network Configuration

​
Local Access Only (Default)

monitor:
  enabled: true
  host: "127.0.0.1"
  port: 8765
Access at: http://127.0.0.1:8765

​
Remote Access

monitor:
  enabled: true
  host: "0.0.0.0"  # Listen on all interfaces
  port: 8765
Access at: http://<server-ip>:8765
Binding to 0.0.0.0 allows external access. Ensure proper firewall rules are in place.

​
Custom Port

monitor:
  enabled: true
  port: 9000
Useful if port 8765 is already in use.

​
Performance Considerations

The monitor adds minimal overhead to the search process:
  • WebSocket updates are asynchronous
  • Solution data is sent in batches
  • AI summaries only run when triggered

​
Optimizing for Long Runs

monitor:
  enabled: true
  max_solution_length: 5000    # Reduce to save memory
  summary_interval: 50         # Less frequent summaries
  summary_top_k: 3             # Fewer programs in summary

​
AI Summary Models

Choose different models based on your needs:

​
Fast & Cheap

monitor:
  summary_model: "gpt-4o-mini"
  summary_interval: 5  # More frequent updates

​
Detailed Analysis

monitor:
  summary_model: "gpt-5"
  summary_interval: 20  # Less frequent, deeper analysis
  summary_top_k: 10

​
Local Model

monitor:
  summary_model: "ollama/llama3"
  summary_api_base: "http://localhost:11434/v1"
  summary_api_key: null

​
Troubleshooting

Error: Address already in useSolution: Change the port number
monitor:
  port: 9000  # Use different port
Issue: Dashboard not loadingChecklist:
  1. Verify monitor is enabled: enabled: true
  2. Check the console for server startup message
  3. Ensure firewall allows connections to the port
  4. Try host: "0.0.0.0" instead of 127.0.0.1
Issue: Summaries not appearingSolutions:
  1. Check API key is set: summary_api_key or OPENAI_API_KEY
  2. Verify model name is correct
  3. Check API base URL
  4. Set summary_interval > 0 for automatic generation
  5. Use manual trigger button in dashboard
Issue: Dashboard consuming too much memorySolutions:
  1. Reduce max_solution_length to 3000-5000
  2. Increase summary_interval to generate summaries less often
  3. Clear old solutions from the dashboard periodically

​
Example Configurations

​
Minimal Monitoring

monitor:
  enabled: true
  # All other defaults

​
Production Monitoring

monitor:
  enabled: true
  port: 8765
  host: "0.0.0.0"
  max_solution_length: 8000
  summary_model: "gpt-4o-mini"
  summary_interval: 15
  summary_top_k: 5

​
Research Setup

monitor:
  enabled: true
  port: 8765
  host: "127.0.0.1"
  max_solution_length: 20000
  summary_model: "gpt-5"
  summary_interval: 0  # Manual only
  summary_top_k: 10

​
Next Steps

Configuration Overview

Learn about other configuration options

Human Feedback

Guide the search with human expertise