Discover and explore top open-source AI tools and projects—updated daily.
EverMind-AISelf-improving agent harness for evolving digital workers
Top 39.6% on SourcePulse
Summary
Raven is a self-improving agent harness built on the EverOS framework, designed to integrate AI agents deeply into daily workflows beyond simple demos. It targets developers and power users seeking durable memory, adaptable skills, and reusable agent templates, enabling agents to continuously refine their capabilities and become more valuable over time.
How It Works
Raven treats the agent's surrounding harness as a core product, moving beyond basic LLM+tools+loop paradigms. Its "memory-first" architecture leverages EverOS for persistent user, agent, and world knowledge, ensuring continuity across sessions. The "self-improving skills" system, SkillForge, identifies and refines repeated workflows into robust, evolving skills. This approach enables proactive agents and facilitates the creation of shareable Agent Templates.
Quick Start & Requirements
Installation is streamlined via a bash script (curl -fsSL https://raven.evermind.ai/install.sh | bash) or PowerShell (irm https://raven.evermind.ai/install.ps1 | iex), handling dependencies like Python 3.12 and Node.js 22. Post-installation, run raven onboard and raven. Raven supports numerous LLM providers (OpenAI, Anthropic, Gemini, etc.) and custom endpoints. raven doctor aids diagnosis. Website: https://raven.evermind.ai.
Highlighted Details
Maintenance & Community
The project encourages community engagement via repository starring. Specific community channels or details on core contributors/sponsorships are not explicitly detailed in the README.
Licensing & Compatibility
Licensed under Apache License 2.0. Agents and templates built with Raven can be freely used, modified, and commercialized under this license, with attribution requested. Some components may originate from MIT-licensed projects.
Limitations & Caveats
Raven is designated "pre-alpha," meaning APIs are subject to change without notice. Components like the Eval engine are only partially implemented, indicating ongoing development and potential instability.
23 hours ago
Inactive