Welcome to Myrm
Myrm is an open-source AI agent workspace: the WebUI, API server, Tauri desktop app, and the agent harness runtime are MIT-licensed; only the optional SaaS control plane is proprietary. You get persistent per-user sandboxes, cross-session memory, and a GUI-first experience with 100+ LLM models, 26 messaging channels, and three deploy modes — local WebUI, Tauri desktop, or SaaS cloud.Why Myrm?
Persistent Sandbox
Your environment survives sleep and wake cycles. Installed tools, files, and configs stay with you across sessions.
Cross-Session Memory
8 memory types with dialectic profiling (style, habits, implicit goals) and intelligent forgetting. Agents remember you across sessions — editable in GUI, not a black box.
GUI-First
Configure agents, manage skills, monitor tasks, and visualize costs from a modern interface. Zero YAML editing required.
Multi-Model
100+ models from OpenAI, Anthropic, Google, DeepSeek, and local providers. Smart routing picks the optimal model per task.
Self-Healing
14-layer error recovery handles model outages, rate limits, and network failures automatically. Agents keep working while you sleep.
Self-Evolving
Agents learn from every interaction. Skills are automatically extracted, tested, and deployed through the evolution engine.
Key Capabilities
Deployment Options
All three modes share the same codebase and feature set. Your data and configuration are portable between modes.
Token Efficiency
Myrm’s turn-one tool disclosure starts at about 1,765 tokens in our benchmark profile (about 2,167 with a fuller toolset) — materially below loading every tool up front. With prompt caching, effective cost can drop further. Treat these as measured defaults for the stated toolset, not a universal guarantee: your enabled tools, skills, and model change the number.What is open
The WebUI, API server, agent harness, and Tauri desktop shell are MIT-licensed. The optional SaaS control plane (multi-tenant billing, LLM relay, fleet management) is proprietary; you do not need it to run the product yourself. That split is deliberate: the part that touches your files, memory, skills, and model keys can be audited, forked, and self-hosted. See Product Architecture for the layer-by-layer contract.Next Steps
Product Architecture
Frontend, server, harness, and desktop — how they connect.
Quick Start
Get up and running in under 2 minutes.
Core Concepts
Learn how Myrm agents work.
Goal Management
Set up autonomous long-running tasks.
MCP Integration
Connect external tools and services.
Migrate from another agent
Scan, preview, import, and roll back — with the honest boundaries.
Competitor comparison
Row-by-row against Hermes, OpenClaw, Codex, Claude Code, and closed Personal Agents.