Model Selection and Session Management

The Sticky Model Picker

At the bottom of the composer, next to where you type your prompts, sits the model picker -- a dropdown that shows your current model and lets you switch to any other model your configured providers support. This picker has two important behaviors that distinguish it from model configuration in the CLI.

First, the model picker is sticky per device. Whatever you select stays selected across restarts, new sessions, and app relaunches. The selection is stored locally on your machine, not in your Hermes profile or config.yaml. This means you can set a preferred model in the Desktop on your laptop without affecting the model your server uses for CLI sessions.

Second, the model picker scopes changes to the current conversation when you switch models mid-chat. If you start a session with Claude Sonnet 4.6, have three turns, and then switch to Gemini 2.5 Pro for the fourth turn, only the fourth turn and subsequent turns use Gemini. The previous turns remain in the conversation history and are sent as context to the new model, but the model switch applies only from that point forward. This lets you do things like: start a task with a cheap model, hit a complex problem that requires more capability, switch to a premium model for that specific step, then optionally switch back.

Reasoning Effort and Fast Mode

For models that support extended reasoning (Claude, for example), the model picker also exposes a reasoning effort control -- typically a slider or preset selection that maps to the model's thinking budget. Higher reasoning effort produces more thorough outputs at the cost of more time and tokens. Lower effort is faster and cheaper but less thorough.

Fast mode (sometimes called Turbo in the Hermes UI) is a preset that selects the fastest-response configuration for the current model. It typically means minimal or no extended reasoning and prioritizes getting a response as quickly as possible. Fast mode is useful for simple tasks where you want quick iteration, not thorough analysis.

Both reasoning effort and fast mode are per-session state remembered in the model picker, not global preferences. Changing them in one session does not change the defaults for new sessions. This prevents you from accidentally locking in a low-effort setting for all future work because you enabled fast mode once.

Session List and Archiving

The left sidebar (below the file browser) shows your session history -- a list of recent conversations, each with a timestamp and an automatically generated title based on the first few messages. As you create more sessions over time, this list can grow long enough to become hard to navigate.

Archiving lets you move sessions you are done with out of the active list without deleting them. Archived sessions are preserved and searchable but do not clutter the main session list. This is useful for project-based workflows where you create one session per task -- when the task is done, archive the session and keep the active list focused on current work.

The session list includes a search field. Type a session ID (available in the session's metadata panel) or keywords from the conversation to find specific sessions quickly. Search works across both active and archived sessions, so archiving is not a barrier to finding old conversations.

Concurrent Sessions Across Profiles

Profiles in Hermes are isolated configuration sets -- separate collections of config, skills, and session history. The Desktop app supports running concurrent sessions in multiple profiles simultaneously. You can have Claude Opus 4 handling a complex code project in your development profile while a lighter model handles customer communications in your business profile, both running at the same time.

Switch between profile sessions in the Desktop using the profile selector in the top-left corner of the app. Each profile maintains its own session list, model selection, and active skills. Switching profiles in the UI changes which profile's sessions and configuration are active in the current window.

For more complex multi-profile workflows, you can open the Desktop app in multiple windows, each attached to a different profile. This requires launching additional instances from the command line with the --isolated flag: hermes desktop --profile work --isolated. Each window then shows only the sessions and configuration for its designated profile.

Cross-Profile Session References

Sometimes work in one profile needs to reference a conversation from another profile. Cross-profile @session links enable this. In a prompt, you can reference a session from a different profile using the session ID prefixed with the profile name:

@session:business/session-abc123

Hermes resolves the reference and makes the referenced session's context available in the current conversation. This is useful for tasks that span profiles -- for example, pulling context from a technical investigation session in your dev profile into a documentation-writing session in your writing profile.

Cross-profile references are read-only: you can read the context of a session in another profile, but you cannot write to it or modify it. This prevents accidental contamination of one profile's sessions by another.

Session Metadata and IDs

Every Hermes session has a unique ID -- a short alphanumeric string that identifies it unambiguously. You can find the ID for the current session in the session metadata panel (accessible by clicking the session title in the header). The ID is what you use for cross-profile references, for finding sessions in search, and for referencing a session from the CLI.

Session IDs are stable: they do not change when you archive a session, rename it, or access it from a different interface. They are the persistent identifier that lets you reliably reference a specific conversation regardless of what else changes around it.