A small server on your machine that hosts your agents, each with its own persona and private memory, a shared long-term wiki they curate while they sleep, and one conversation that flows across Claude, ChatGPT and any local model.
Get somora running on your own machine.
Lab // Experimental · Please read this before installing somora.Agents, memory, models and tools, working together in one persistent system.
Create specialized agents that collaborate, delegate tasks and call on each other when needed.
Agents remember across sessions and turn individual experience into knowledge the whole team can use.
somora turns conversations into memory and memory into lasting knowledge through autonomous background dream cycles.
Switch between cloud and local models without losing your conversation, memory or tools.
Connect files, terminals, browsers, MCP servers and the tools your agents need to get actual work done.
Dedicated coding agents can plan, build, run and test software, directly or on behalf of other agents.
Agents can schedule future work, wake themselves up and continue tasks autonomously with Sentinel.
Run somora on infrastructure you control, with local models, local memory and your own tools.
somora is used daily and keeps growing. A few of the things waiting past the first eight:
Web desktop with multi-window chat · Mobile PWA · Terminal TUI · Wiki Explorer with link graph · Voice in & out · Image & video generation · Skills from ClawHub · Sub-agents & agent-to-agent · tmux sessions · SSH resources · Projects · External MCP servers · Obsidian vault recall · 40+ typed tools · Per-agent abilities · Session compaction · Tailscale HTTPS · HTTP + SSE + WS API
Explore the docs →It grew out of countless hours of exploring what happens when agents, models, memory and tools stop being separate pieces and begin working together as one system.
somora is primarily built in Klosterneuburg, Austria, by 👨💻 Luca and 👨💻 Rene, together with their agent teams led by 🤖 Donna and 🤖 Naxon. Yes, the agents building somora are running on somora. Along the way, they're joined by contributors on GitHub, with additional resources and support from enovom.
We use somora ourselves every day. We build with it, break it, change it, and occasionally discover that it has ideas of its own. That's part of the experiment.
And experimental is exactly how we consider its current state.
somora is not production-ready software, and we strongly recommend treating it accordingly. An agent harness can execute tools, access files, run commands and interact with other systems. Given the wrong instructions, configuration, permissions or simply an unexpected model response, it can cause real damage.
There are no guarantees. Things may break, change or disappear. Use somora at your own risk, and only give agents access to systems and data you can afford to lose.
If that doesn't scare you away, welcome to the experiment.
We'd love to hear what you think. Share your feedback, ideas, bug reports or experiments with us on GitHub, join the conversation on Discord, or simply send us an email at [email protected].