無程式碼也能輕鬆打造專業LINE官方帳號！一鍵導入模板，讓AI助你行銷加分！
Imagine an AI agent that not only executes tasks but also reviews its own performance, identifies its weaknesses, and modifies its own source code to do better next time. That is precisely what Evolver delivers. Built by the Chinese AI team EvoMap (evomap.ai), Evolver is an open-source self-evolution engine for AI agents, powered by the Genome Evolution Protocol (GEP).
With over 4,200 GitHub stars and connections to 130,000+ AI agent nodes processing 46 million cumulative calls, Evolver represents a significant step toward truly autonomous AI systems. It operationalizes a concept that has long been theoretical: agents that improve themselves through experience, much like biological organisms evolve through natural selection.
graph TD
A[Agent Execution] --> B[Log Collection]
B --> C[Signal Extraction]
C --> D{Gene/Capsule Selection}
D --> E[GEP Prompt Generation]
E --> F[Code Modification]
F --> G[Test Validation]
G -->|Pass| H[Solidification]
G -->|Fail| D
H --> I[Publish to Network]
I --> AWhat Is Evolver?
Evolver is a self-evolution engine designed to sit alongside an AI agent’s runtime. It monitors agent execution logs, extracts meaningful signals about errors, inefficiencies, or suboptimal behaviors, and then generates structured prompts that cause the agent to modify its own code. The result is an agent that demonstrably improves over time without human intervention.
The engine is configured to protect its own source files – it can modify the agent’s code and behaviors, but it cannot modify Evolver itself. This architectural boundary ensures that the evolution mechanism remains stable even as the agent evolves.
What Is the Genome Evolution Protocol (GEP)?
The Genome Evolution Protocol is the structured specification that makes Evolver’s self-evolution possible. It draws inspiration from biological genetics:
| Concept | Biological Analogy | Evolver Implementation |
|---|---|---|
| Gene | Atomic trait | Atomic capability unit (e.g., “read file”, “execute SQL query”) |
| Capsule | Successful survival strategy | A saved successful task execution path, packaged for reuse |
| EvolutionEvent | Mutation record | Immutable log entry with SHA-256 hash, tracing every change |
| Genome | Complete DNA | The full set of Genes and Capsules defining an agent’s capabilities |
Each evolution event is content-addressed using SHA-256 hashes, which provides deduplication and tamper-proofing. The protocol is model-agnostic, meaning GEP assets work across different LLM backends including GPT-4, Claude, and open-source models.
How Does Self-Evolution Work?
Evolver implements a rigorous 10-step evolution cycle that ensures changes are meaningful, tested, and traceable:
| Step | Action | Description |
|---|---|---|
| 1 | Scan Logs | Read agent execution memory and history |
| 2 | Extract Signals | Identify errors, inefficiencies, and patterns |
| 3 | Select Gene/Capsule | Choose which capability or strategy to evolve |
| 4 | Generate GEP Prompt | Create a structured evolution directive |
| 5 | Execute Modification | Apply the change to the agent’s codebase |
| 6 | Validate with Tests | Run automated tests to verify correctness |
| 7 | Solidify | Lock in successful changes |
| 8 | Publish to Network | Share the evolved capability across the agent network |
| 9 | Log EvolutionEvent | Record the immutable change record |
| 10 | Return to Monitor | Begin the next evolution cycle |
Evolver supports four configurable evolution strategies:
- balanced – Moderates between innovation and stability
- innovate – Prioritizes novel approaches and aggressive optimization
- harden – Focuses on error resistance and edge case handling
- repair-only – Only fixes identified bugs, no behavioral changes
Signal de-duplication prevents repair loops, ensuring the agent does not repeatedly attempt to fix the same issue.
The Hermes Agent Controversy
In April 2026, Evolver became the center of a notable open-source controversy when EvoMap publicly accused Nous Research’s Hermes Agent of copying Evolver’s self-evolution architecture. The accusation pointed to the same 10-step evolution loop with 12 term substitutions and zero attribution. Hermes Agent denied the claims, citing earlier repository creation dates.
The incident sparked significant debate in the AI community about attribution in open-source AI development, particularly as self-evolving agents become more common. It also highlighted Evolver’s role as a pioneering implementation in this space, regardless of one’s position on the specific dispute.
Getting Started with Evolver
Evolver is easy to set up and run:
git clone https://github.com/EvoMap/evolver.git
cd evolver
npm install
node index.js # Single evolution run
node index.js --loop # Continuous evolution daemon
Prerequisites are minimal: Node.js 18 or higher and Git. The engine is offline-capable for its core functionality, making it suitable for air-gapped or secure environments. Configuration options include the evolution strategy, signal thresholds, test framework integration, and network publishing settings.
FAQ
What is Evolver? Evolver is an open-source GEP-powered self-evolution engine for AI agents. It enables agents to analyze their own execution logs, detect errors and inefficiencies, and autonomously generate improvement prompts that modify their own source code.
What is GEP? GEP stands for Genome Evolution Protocol. It is a structured specification for encoding agent capabilities as Genes (atomic units), packaging successful execution paths as Capsules, and recording all changes as immutable EvolutionEvents.
How does self-evolution work in Evolver? Evolver follows a 10-step cycle: scan logs, extract signals, select a Gene or Capsule, generate a GEP prompt, execute the modification, validate with tests, solidify the change, and publish to the network. Each mutation is gated by an explicit Mutation object.
How do I install Evolver? Clone the repo from github.com/EvoMap/evolver, run npm install, then run node index.js for a single evolution run or node index.js –loop for continuous evolution mode. Requires Node.js 18 or higher.
Does Evolver work offline? Yes, Evolver’s core functionality is offline-capable. GEP assets use content-addressed SHA-256 IDs for deduplication and tamper-proofing, and the evolution engine does not require internet connectivity for basic operation.
