AI Personality

UnisonOS makes assistant personality and priorities person-specific, portable, and independent from any specific model.

See also: Default System Prompt (Canonical).

Why Separate Model from Identity?

Models change. Hardware changes. Providers change.

If personality lives inside a model configuration, a person loses continuity during upgrades and replacements. UnisonOS instead stores: - persistent identity and values - mutable priorities and directives outside the model, then compiles and injects the active system prompt at runtime.

Customize the Assistant

Option A: Edit Files Directly

Edit: - ~/.unison/prompt/user/identity.json (the person’s identity and values) - ~/.unison/prompt/user/priorities.json (the person’s priorities and directives)

Changes apply on the next turn by default.

Option B: Ask the Assistant to Propose a Change

Example request:

“Update your priorities to be more concise and challenge my assumptions.”

UnisonOS will: 1. propose a patch 2. validate it against schemas 3. require approval for high-risk changes 4. apply it, snapshot, and log the change

Roll Back

If you don’t like a change, roll back to a prior snapshot in: ~/.unison/prompt/snapshots/

Anti-Sycophancy

UnisonOS explicitly enforces: - disagreement when justified - tradeoffs over reassurance - correction of factual errors - minimal flattery

This is configurable via identity.json (see anti_sycophancy.challenge_level).