Hermes, with a memory that outlives the process

Hermes, with a memory that outlives the process.

The other open harness, same memory, same tools, one token.

Hermes is the second open harness gbrain supports directly, alongside OpenClaw. Both are covered in the open-source repo, and both connect the same way.

What it gains is what any harness gains here: a brain that keeps what it reads, and a set of tools it can reach without a key sitting in a config file. Mail, calendar, web search and page fetching arrive as one MCP server you scope yourself.

Swapping harness does not cost you the memory. It is the workspace’s, not the agent’s.

One endpoint connects Hermes

Not one per service. What this client can call is decided by its permission, not by the address, and the token comes from the client's own page in the workspace, minted once and shown once.

Endpoint: https://gbrain.io/mcp
Header: Authorization: Bearer <token>

Every application Hermes gets

Connect one in the browser and it answers here, under the same permission and the same record as everywhere else.

Gmail Search, read, file and draft mail from a terminal or an assistant.
Google Calendar List, add, change and cancel events without leaving the terminal.
Web search Search the live web and get back text, not a page of links.
Page fetching Turn a URL into readable text, including pages that need a browser.
Google Drive Not yet Read and search the documents your team keeps in Drive.
Company research Not yet Company bios and investor searches, from the terminal or any assistant.

The other places a connection answers are on Works with , and Tools explains the permission, the ceiling and the record they share.