Skip to content
vavio

Can two Hermes profiles share a model account without sharing conversations?

Yes. Two Hermes Agent profiles on the same computer can use the same model account, and each keeps its own conversations, memory, skills, SOUL.md, cron jobs and settings. Profiles separate Hermes’s state; they are not a security boundary. On the default local terminal backend, both profiles run as the same operating-system user, can reach the same files, and by default share command-line logins such as git, ssh and gh. One profile can even read another’s conversation history if asked to. Use profiles to keep work tidy. Use separate computers when one agent must not be able to see the other’s data.

This guide follows Hermes v0.23.1, commit 3b42157, checked on 13 November 2026. We reviewed the official source and documentation. We did not run a multi-profile trial on a Vavio computer for this article.

What does a profile keep separate?

A profile is its own Hermes home directory. The default profile is ~/.hermes; a named profile such as work lives at ~/.hermes/profiles/work. Hermes sets HERMES_HOME to that directory when you run hermes -p work or the work command alias, and state is read from and written to that home.

Kept per profile Shared by profiles on one computer
Conversation history (state.db) and session search by default The Hermes installation and its updates
Built-in memory (MEMORY.md, USER.md) Your operating-system account and its files, unless you sandbox the terminal
Skills, SOUL.md, config.yaml and model choice Command-line logins in your real home folder, unless terminal.home_mode: profile is set
API keys in that profile’s .env Model credentials added with hermes auth in the default profile, and keys exported in your shell, unless the profile adds its own for that provider
Cron jobs and messaging bot tokens Python plugin dependencies
Logs and gateway state Any external memory provider pointed at the same store

Upstream warns against pointing two running agents at one profile: both write memory automatically, and each loads the other’s writes at the next session start. A separate profile per agent avoids that.

How do two profiles share one model account?

How you share the account depends on how you sign in.

With an API key, such as an OpenRouter key, each profile reads its own .env file. A named profile does not read the default profile’s .env. Put the key in both files, or create the second profile with hermes profile create work --clone, which copies config.yaml, .env, SOUL.md, skills and the two curated memory files, but no sessions or cron jobs. Two other routes reach every profile: a key you added with hermes auth add in the default profile, which a named profile uses for any provider it has no credential of its own for, and a key exported in your shell. If a profile must not spend on a key, give it its own key for that provider and don’t export the shared one.

With an OAuth login, such as a Claude Pro/Max, OpenAI Codex or xAI sign-in, a named profile with no login of its own for that provider falls back to the default profile’s auth.json. Hermes does not copy these logins into a --clone-all copy, because the refresh token can only be used once. A copy would be the same credential with two owners, and the first refresh would revoke it for the other. When a profile refreshes a borrowed login, it writes the new token back to the default profile, so both stay signed in. To give a profile its own login, run hermes -p work auth add <provider> inside it; that profile then uses its own login instead.

Either way, one account means one bill and one set of provider limits. hermes insights reads each profile’s own session history, so you can compare the two profiles’ token use, but the provider sees one customer. If you need a hard cap or a clean bill per agent, use a separate key for each profile where your provider supports one. OpenRouter, for example, recommends that each agent key get its own credit limit.

Can one profile see the other’s conversations?

Not by default, but it can if asked. A session search without a profile name reads only the current profile’s state.db, and a bare session ID from another profile returns “not found in this profile”. The session_search tool also accepts a profile argument, however, and with it opens another profile’s database read-only. That is how an @session:<profile>/<id> link resolves. So the separation keeps conversations from mixing; it doesn’t stop one profile from looking.

The same applies to files. Upstream states that profiles do not sandbox the agent: on the default local terminal backend, Hermes has the same filesystem access as your user account. Hermes’s file tools refuse to read files such as .env, but the terminal runs with your account’s permissions. A profile’s SOUL.md can tell the model to stay within one project; it cannot enforce it.

What else should I change for a clean split?

  • Working folder. Set an absolute terminal.cwd in each profile’s config.yaml. Otherwise CLI commands start where you launched Hermes, and a messaging gateway starts in your home folder.
  • Command-line identities. Set terminal.home_mode: profile if each agent needs its own git, gh, cloud or ssh login. Hermes then gives tool commands HOME={HERMES_HOME}/home, and you set up those logins inside it.
  • Messaging bots. Give each profile its own Telegram, Discord or Slack bot token. Clones leave channel settings behind unless you pass --clone-channels, and on Telegram, Discord, Slack, WhatsApp and Signal a second standalone gateway using the same token is blocked with an error.
  • Memory. --clone copies the curated memory files at creation; after that, the two drift apart. For a blank start, create the profile without --clone. If you use an external memory provider, check that each profile writes to its own store.
  • Scheduled work. Cron jobs belong to the profile that created them and are not cloned, so a job never runs twice by accident.

When are profiles enough?

Profiles suit one person who wants two tidy agents: a personal assistant on one bot and a research agent on another, each with its own memory, instructions and schedule, both paid from one model account. They don’t suit two people who must not see each other’s data, an agent that handles credentials the other mustn’t reach, or a test agent that could break production. For those, run Hermes on separate computers or under separate operating-system users.

Using profiles on Vavio

Vavio provides a managed Hermes computer for $14.99 per month, with model usage billed separately. Its MCP server and CLI let a supported assistant connect a model, check Hermes and recover it. With the owner’s explicit permission for these reads, those controls can list a computer’s agents and read each agent’s conversations, model choices, logs and key settings, never credentials. They don’t create agents. Extra agents on one Vavio computer are standard Hermes profiles that you create with the commands above, and we haven’t tested every multi-agent setup for customers yet. To pass work from one agent to another, see how agents hand work over on the Kanban board. If you need two agents that are fully separate today, use one computer for each, and compare the current hosting offer. If you self-host, the profile commands above work as described in the official profiles guide.

Sources

Written by Vavio with AI assistance and checked against pinned official Hermes v0.23.1 source and documentation. This is a source review, not a live multi-profile trial.

More from the Vavio blog: Hermes Agent guides