Hermes Workspace ships v1.0

Most people use an artificial intelligence agent like a mechanic working out of five open toolboxes: one tab for chat, one terminal window, one file tree, one log pane, and one settings page. Hermes Workspace v1.0 tries to pull those pieces into one screen, and its GitHub repository showed the v1.0.0 release commit on April 11, 2026. (github.com) That matters because Hermes itself is not just a chatbot. Nous Research describes Hermes Agent as a long-running system with memory, reusable skills, six terminal backends, and support for messaging platforms like Telegram, Discord, Slack, WhatsApp, Signal, and email. (hermes-agent.nousresearch.com) A workspace in this context is basically a cockpit for that agent. Instead of reading raw logs and typing commands in separate windows, Hermes Workspace shows chat, files, memory, terminal access, and tool activity in one interface. (hermes-workspace.com) The new part in v1.0 is that the workspace is no longer just a thin chat wrapper around one model provider. The project’s latest commit mentions profiles, a knowledge browser, Model Context Protocol settings, and a skills hub, which turns it into a control layer for switching how an agent runs. (github.com) Profiles are the practical piece here. The upstream Hermes project recently added gateway behavior where the application programming interface server model name is derived from the profile name, and Hermes Workspace v1.0 adds profile management on the front end so one setup can point different jobs at different model stacks. (github.com, github.com) That is why the product page talks about Claude, Generative Pre-trained Transformer models, Gemini, Codex, and local models in one conversation flow. Hermes Workspace is built to sit on top of any OpenAI-compatible backend, so the same interface can talk to a cloud model or a local model server without changing the whole workflow. (hermes-workspace.com, github.com) Local model support matters because tools like Ollama let people run models on their own hardware. Hermes Workspace does not ship its own model, but it can connect to local or self-hosted backends through the standard chat completions interface, which is the software equivalent of plugging different appliances into the same wall socket. (github.com) The “skills” piece is Hermes’s version of saved procedures. Nous Research says Hermes can create and improve skills over time, and Hermes Workspace adds a browser for those skills so users can inspect, launch, and manage them from the same place they watch the agent work. (hermes-agent.nousresearch.com, hermes-workspace.com) The marketplace angle comes from that skills hub. The workspace site says users can browse 100-plus skills inside the interface, while the wider Hermes ecosystem is documented as compatible with open standard skills, which pushes the product closer to an app store for agent behaviors than a normal chat box. (hermes-workspace.com, hermes-agent.nousresearch.com) There is also a second bet inside the release: the terminal stays in the product. Hermes Agent can run commands on a local machine, in Docker, over Secure Shell, or on cloud backends like Modal and Daytona, and Hermes Workspace keeps a full terminal pane next to the chat so the agent and the human can work in the same environment instead of handing work back and forth. (hermes-agent.nousresearch.com, hermes-workspace.com) So v1.0 is less about a prettier interface and more about where the artificial intelligence tooling market is drifting. The winning products increasingly look like operating desks for many models, many tools, and many long-running agents, and Hermes Workspace is positioning itself as the open-source control surface for that setup rather than another single-model chat app. (github.com, hermes-workspace.com, hermes-agent.nousresearch.com)