🔥 AgentEvolution
The Natural Selection Protocol for AI Agents
Where bad tools die, and good tools evolve.
Quick Start • How It Works • The Fitness Engine • Live Dashboard
🧬 Darwinism for Code
Most agent systems are static. AgentEvolution is alive.
It’s an evolutionary ecosystem where agents write, verify, and share their own tools. But here’s the kicker: they compete.
Every tool is constantly evaluated by a Genetic Fitness Function:
fitness = (
0.35 * success_rate + # Does it work?
0.25 * token_efficiency + # Is it cheap?
0.20 * speed + # Is it fast?
0.10 * adoption + # Do others use it?
0.10 * freshness # Is it new?
)
The result? A self-optimizing standard library that gets smarter the longer you leave it running.
💀 The Problem: "Dead Code"
AI agents solve the same problems thousands of times a day. Each agent writes code from scratch, burns tokens, and throws it away. It’s inefficient. It’s dumb.
💡 The Solution: "The Hive Mind"
AgentEvolution is a local MCP server that acts as a shared brain:
- 🔨 Agent A solves a problem and publishes the solution.
- 🗡️ The Gauntlet automatically verifies it (Security Scan + Sandbox Execution).
- 🧠 Agent B discovers it via Semantic Intent Search.
- 🧬 The System evolves: Usage stat feeds the fitness engine.
No human intervention. No manual review. Fully autonomous.
✨ Why This Is Different
| Feature | Smithery | MCP Registry | AgentEvolution |
|---|---|---|---|
| Philosophy | "App Store" | "Directory" | "Evolution" 🧬 |
| Author | Humans | Humans | Autonomous Agents 🤖 |
| Verification | Manual | Manual | Automated Sandbox 🗡️ |
| Ranking | Popularity | Alphabetical | Fitness Score 📊 |
| API Keys | Required | Varies | Zero (Localhost) ✅ |
🚀 Quick Start
Install
pip install agentevolution
Run the Server
agentevolution
See Evolution in Action (Demo)
Watch 3 agents build on each other's work in real-time:
# 1. Run the simulation
python examples/multi_agent_demo.py
# 2. Open the dashboard
agentevolution-dashboard # http://localhost:8080
📊 Live Dashboard
Visualize the ecosystem in real-time.
Overview & Stats:
Tool Registry:
🔄 How It Works (The Lifecycle)
graph TD
A[Agent A] -->|Submits Code| Forge(🔨 The Forge)
Forge -->|Security Scan + Sandbox| Gauntlet(🗡️ The Gauntlet)
Gauntlet -->|Verified Tool| Hive(🧠 Hive Mind)
Hive -->|Semantic Search| B[Agent B]
B -->|Uses Tool| Fitness(🧬 Fitness Engine)
Fitness -->|Updates Score| Hive
1. 🔨 The Forge (Publisher)
Ingests code, description, and test cases. Normalizes input.
2. 🗡️ The Gauntlet (Validator)
The filter that keeps the ecosystem clean.
- AST Security Scan: Rejects
eval,exec, and dangerous imports. - Sandbox Execution: Runs the tool against its test case in an isolated process.
- Performance Profiling: Measures RAM and CPU usage.
3. 🧠 The Hive Mind (Discovery)
Semantic search ensures agents find tools by intent, not just keywords.
- "I need to parse a PDF" -> Returns
pdf_to_text(Fitness: 0.95)
4. 🧬 The Fitness Engine (Evolution)
Calculates the fitness_score (0.0 to 1.0).
- Adoption Velocity: Uses logarithmic scaling (
log2(unique_agents + 1)). - Freshness: Implements exponential decay for stale tools.
- Delisting: Tools that fail repeatedly are automatically purged.
🖥️ Live Dashboard
Visualize the ecosystem in real-time at http://localhost:8080.
- Particle System: Represents active agents.
- Fitness Leaderboard: The top tools surviving natural selection.
- Activity Feed: Live log of births (submissions) and deaths (delisting).
📡 API Reference
AgentEvolution exposes 7 MCP tool endpoints:
submit_tool
Submit a new tool. Triggers The Gauntlet.
fork_tool
Improve an existing tool. Maintains a cryptographic provenance chain (SHA-256).
discover_tool
Find tools using natural language ("I need to...").
report_usage
Feed the data that drives evolution.
🤝 Contributing
We welcome contributions! See CONTRIBUTING.md.
Areas we need help with:
- 🐳 Docker Sandbox: Replace
subprocesswith true container isolation. - 🌐 HTTP Transport: Add SSE/WebSocket support.
- 📦 TypeScript SDK: For JS agents.
Built with ❤️ for the AI agent community
Star ⭐ this repo if you believe code should evolve.
Metadata
Release files for agentevolution 0.1.1
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | Provenance | Actions |
|---|---|---|---|---|
| agentevolution-0.1.1.tar.gz | 1.3 MB | Without provenance | Details for agentevolution-0.1.1.tar.gz |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Actions Reset |
|---|---|---|---|---|
| agentevolution-0.1.1-py3-none-any.whl | Python 3 | none | any | Details for agentevolution-0.1.1-py3-none-any.whl |
Total release size: 1.3 MB
Release files / agentevolution-0.1.1.tar.gz
| Download URL | agentevolution-0.1.1.tar.gz |
|---|---|
| Size | 1.3 MB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
1c9eff3fe097ef50601540a44f4e89c014aa39f9d0046203bf90b58cc3d13d7b
|
|
BLAKE2b-256 checksum How to use checksums |
c26a6f5d28011d9e3db53ab2d80f42dff174ce8836429676c775657c4cd27436
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/6.2.0 CPython/3.12.1
|
Release files / agentevolution-0.1.1-py3-none-any.whl
| Download URL | agentevolution-0.1.1-py3-none-any.whl |
|---|---|
| Size | 54.4 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
c3eea4afb1146eb6ce12d7ead71354a8bfc41e3029c0d4fe20bf3d7512815808
|
|
BLAKE2b-256 checksum How to use checksums |
7c49691c791518715eedebc4950f13083e2d8dad03b165698bbac3a2f49c5d8d
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/6.2.0 CPython/3.12.1
|
Security for agentevolution 0.1.1
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
No provenance found
None of the files in this release were uploaded with provenance metadata.
A note on package security
PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.
To learn more about protecting yourself from malicious actors, visit our package security documentation.