Persistent memory
Completed work becomes a reusable skill. The tenth time a task comes round it is a replay, not a fresh chain of reasoning — which means it is faster, cheaper and more consistent.
01 / What it is
An agent that keeps what it learns.
Hermes Agent is an open-source, MIT-licensed self-hosted AI agent from Nous Research, released in February 2026. Its defining feature is persistent memory: after finishing a task it can save that work as a skill and reuse it later, instead of reasoning the same problem from scratch every time.
It is not our project. It is open source, and Nexum's work is deploying, configuring and operating it on your behalf.
Where your team already is
Reachable on Telegram, Discord, Slack, WhatsApp, Signal, email and the command line — no new app for anyone to remember to open.
Model agnostic
Runs on Claude, GPT, Gemini, DeepSeek, Qwen or any OpenAI-compatible endpoint, so a change of provider is a config change rather than a rebuild.
Hermes Agent
MIT licensing and self-hosting mean no per-seat lock-in and no obligation to send your work through someone else's service. The agent, its memory and its credentials sit on infrastructure you own.
Reaches
Runs on
02 / What Nexum does
Deployed, scoped and looked after.
Same engagement as our other agent work: we stand it up on your infrastructure, connect it to the channels and tools you use, shape it around the tasks you repeat, and keep it healthy as models and requirements move.
Deploy
Self-hosted on your servers or cloud tenancy, with memory storage configured to
your retention rules.
Connect
The messaging channels your team actually uses, plus the tools and accounts the
agent needs — and nothing beyond them.
Teach
Seed the first skills against your real workflows so the memory is useful from
week one rather than month six.
Maintain
Upgrades, model changes, memory review and monitoring, handled on an ongoing
basis.