A dynamic learning framework that enables AI agents to accumulate experience, verify outcomes, and self-correct through a persistent memory database.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install live-evo
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install live-evo using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is Live-Evo: Self-Evolving AI Memory System?
Live-Evo is a specialized memory architecture designed to help AI agents evolve by learning from every task they perform. Unlike static instruction sets, this system actively retrieves relevant past experiences to generate task-specific guidelines, ensuring that the agent does not repeat previous mistakes. By integrating this into your Openclaw Skills workflow, you create a feedback loop where successes and failures directly inform future performance.
The system utilizes a persistent JSONL database to store granular details about task attempts, including failure reasons and key improvements. It allows agents to operate with higher autonomy by providing a framework for contrastive evaluation—comparing outcomes with and without memory—to scientifically improve the quality of generated solutions over time.
Live-Evo: Self-Evolving AI Memory System Use Cases
- Automated software debugging where test results provide a clear signal for learning from failed bug fixes.
- Complex data analysis workflows that require consistent application of past insights and methodologies.
- Predictive modeling tasks where the accuracy of previous forecasts can be used to refine current logic.
- High-stakes coding implementations that benefit from a dual-attempt verification process before finalization.
- Interactive sessions where user corrections serve as primary learning signals for behavior modification.
How Live-Evo: Self-Evolving AI Memory System Works
- Retrieve relevant experiences from the local database by querying the current task description to find similar historical contexts.
- Compile these experiences into a task-specific guideline that synthesizes past lessons into actionable advice.
- Evaluate whether to perform a contrastive attempt (Attempt A vs. Attempt B) or a direct application based on the complexity and cost of the task.
- Execute the task using the provided guidelines while monitoring environment signals like shell output or test failures.
- Log the outcome by adding a new experience entry or updating the weights of existing memories to reflect their actual utility.
Live-Evo: Self-Evolving AI Memory System Setup
To begin using this memory system within your Openclaw Skills, ensure the python scripts are accessible in your environment. Experience data is automatically managed in a persistent directory.
# Check the current status of the memory database
python scripts/stats.py
# Manually retrieve guidelines for a specific task
python scripts/retrieve.py --query "implementing oauth2 in fast-api"
# List all stored experiences to review the learning history
python scripts/list_experiences.py
Live-Evo: Self-Evolving AI Memory System Data Schema & Taxonomy
The system maintains a structured repository of knowledge within the ~/.live-evo/experience_db.jsonl file. Each entry follows a specific taxonomy to ensure high-fidelity retrieval:
| Attribute | Description |
|---|---|
| question | The specific task or problem the agent was asked to solve. |
| failure_reason | A technical breakdown of why a previous approach was unsuccessful. |
| improvement | The specific lesson or strategy that led to a better outcome. |
| weight | A dynamic score (0.1 - 2.0) that determines the relevance of the experience. |
| category | Classification tags such as coding, analysis, or debugging. |
Live-Evo: Self-Evolving AI Memory System Advanced Features
- Contrastive Evaluation Logic: Allows the agent to run side-by-side tests to verify the efficacy of its own memory.
- Dynamic Weight Decay: Automatically prioritizes high-quality, successful experiences while phasing out less relevant or low-weight entries.
- Feedback-Driven Learning: Capable of ingesting direct user corrections and environment errors as primary training data.
- Task-Specific Guideline Synthesis: Moves beyond simple retrieval by merging multiple past lessons into a single, coherent strategy for the current task.
SKILL.md
Loading
Related Openclaw Skills
MerchantGuard
MerchantGuard is a comprehensive security and compliance layer for AI agents, providing scanning, adversarial testing, and certification for payment-handling systems.
Image to PDF Merger Skill
A sophisticated tool for merging multiple images into professional PDF documents with automated orientation correction and OCR capabilities.
OpenClaw Backup Skill
A streamlined utility for creating timestamped backups of critical OpenClaw configuration files, agents, and credentials.
Lyria Music Generation
A high-performance AI music generation skill that creates 30-second instrumental tracks using Google Lyria.
Hello World Skill
A fundamental greeting skill for AI agents that provides personalized, multilingual messages with dynamic date and time integration.
SerpAPI MCP Search
A sophisticated search integration that connects AI agents to SerpAPI via the Model Context Protocol for real-time, structured web data.
Featured*
