anvie/evonic

What it solves

Evonic is an agentic AI framework designed to move agents from concept to production. It solves the complexity of managing multi-agent systems by providing a unified way to define agent identities, tools, knowledge bases, and execution environments, while ensuring safety and security through real-time activity monitoring.

How it works

Evonic uses a modular architecture where agents are defined by six dimensions: concept, model, tools, knowledge base, channels, and skills. It employs a hybrid memory engine (Evomem) combining semantic vector search with a knowledge graph for long-term memory. To ensure security, it uses an "Injection Guard" for prompt injection detection and a heuristic mal-activity detection system that blocks dangerous actions (like mass file deletion) before they execute.

Who it’s for

It is built for developers and organizations looking to build autonomous AI agents, multi-agent swarms, or automated DevOps and research assistants that can operate across local and remote infrastructure.

Highlights

• Workplaces: Execution environments that can be local, remote (via SSH), or tunnel-based (via Evonet) for distributed execution.
• Agent-to-Agent Communication: Native protocols for messaging, delegation, and hierarchical orchestration between agents.
• Knowledge Graph: An interactive visualization of KB documents with Obsidian-style wiki-links and an autonomous KB Organizer sub-agent for entity extraction and deduplication.
• Training Data Archive: Captures byte-exact LLM I/O (including chain-of-thought) to create SFT datasets from real agent runs.
• Multi-Channel Support: Native integrations for Telegram, WhatsApp, and Discord.
• Token Compression: Uses RTK-based compression to reduce LLM costs without losing meaning.