Categories
Self-Hosted AI Agents Workflow Automation Knowledge Graph AI Memory Multi-Agent Orchestration Agentic AI Platform LLM FrameworkStars
Forks
Watchers
Developer links
Evonic
AI agents that delegate to each other, remember past conversations, and stay inside your safety rules all live in one self-hosted platform called Evonic. Every agent in Evonic has its own identity, model, tools, knowledge base, channels, and skills, and a guided first-run wizard creates the first one in minutes against OpenRouter, Together AI, Ollama, DeepSeek, llama.cpp, or any custom OpenAI-compatible endpoint. Agents message and delegate to each other, run commands in a Docker, Podman, or bubblewrap sandbox, and execute on a local folder, an SSH server, or a tunnel device that needs no public IP. A built-in Kanban board hands tasks to agents, a scheduler fires cron jobs and reminders, and Telegram, WhatsApp, and Discord channels put agents where users already chat. Markdown knowledge bases with wiki-links form an interactive graph, a memory engine recalls facts across sessions, and an organizer sub-agent extracts entities and merges duplicates. Safety layers inspect actions for mass deletion or remote code execution and escalate to a human, and an injection guard blocks prompt manipulation. Evaluation dashboards score models across domains, a token monitor tracks usage per agent and model, and bundled skills, 12 plugins, backup and restore, and themed light and dark modes round it out. Running on a dedicated VPS on RepoCloud with guaranteed CPU, RAM, and SSD, full root SSH access, and a browser serial console. AGPL-3.0 licensed.
Benefits
- Agents That Work As A Team
- Agents message, delegate to, and coordinate with one another, so a supervisor agent can hand work to specialists and a Kanban board tracks every task.
- Run Commands Safely Anywhere
- Tools execute in a Docker, Podman, or bubblewrap sandbox on a local folder, an SSH host, or a tunnel device, with heuristic detection stopping dangerous actions.
- Use Any Model Endpoint
- Connect OpenRouter, Together AI, Ollama, DeepSeek, llama.cpp, or any OpenAI-compatible API, and compare candidates with the built-in evaluation engine and token monitor.
- Memory That Persists
- Agents keep long-term memory and a linked markdown knowledge base, so facts, people, and decisions from earlier conversations are recalled instead of repeated every session.
Features
- Setup Wizard And Dashboard
- A four-step wizard picks the provider, creates the first agent, and sets the admin password, then a dashboard summarizes agents, tools, schedules, and token usage.
- Skills And Plugins
- Toggle bundled installable skills and 12 event-driven plugins, search them, or upload your own packages, including MCP client, GitHub webhook, and OpenAI-compatible agent API plugins.
- Chat Channels
- Connect each agent to Telegram, WhatsApp, or Discord from its Channels tab, and add scheduled prompts, reminders, and cron-based triggers for recurring work.
- Knowledge Graph
- Browse a force-directed graph of markdown documents joined by wiki-links, filter nodes by type, and let the organizer sub-agent merge duplicates and repair dangling links.
- Evaluation Dashboard
- Run scored LLM evaluations across domains and difficulty levels with custom evaluators, review results over time, and watch real-time logs as a run executes.
- System Settings
- Search a categorized settings panel for models, routing, safety, users, shared channels, and logs, and pick from dark and light color themes.