Evolve
by agenttoolkit
Self-improving agent framework that learns from conversation trajectories to generate persistent guidelines and best practices.
What it does\nEvolve allows AI agents to learn 'on-the-job' by analyzing their own interaction trajectories. It captures successful patterns and converts them into guidelines and policies stored in a vector database (Milvus), which the agent can then retrieve to improve future performance on similar tasks.\n\n## Tools\n- save_trajectory: Analyzes a conversation trace and automatically generates new guidelines based on successful outcomes.\n- get_guidelines: Retrieves relevant, learned best practices for a specific task context.\n- create_entity: Manually adds a guideline or policy to the knowledge base with optional conflict resolution.\n- delete_entity: Removes an outdated or incorrect guideline by ID.\n\n## Installation\nUsing uv (recommended):\njson\n{\n "mcpServers": {\n "evolve": {\n "command": "uv",\n "args": ["run", "evolve-mcp"]\n }\n }\n}\n\n\n## Supported hosts\n- Claude Desktop\n- Claude Code
Quick install
uv run evolve-mcpInformation
- Pricing
- free
- Published
- 4/12/2026
- stars







