Personas

Each agent has a persona defined through four workspace files. These files shape how the agent thinks, communicates, and works.

IDENTITY.md

Defines who the agent is:

Name — the agent’s display name
Creature — what kind of entity it is (e.g., “Robot”, “Research specialist”)
Vibe — communication style (e.g., “Sharp, calm, warm”)
Emoji — avatar emoji
Avatar — avatar image URL

SOUL.md

Defines how the agent behaves:

Core Truths — fundamental principles the agent follows
- Example: “Be genuinely helpful. Be resourceful before asking. Earn trust through competence.”
Boundaries — what the agent will not do
- Example: “Do not fabricate results. Do not hide uncertainty.”
Vibe — tone and communication style
- Example: “Concise when needed, thorough when it matters.”
Continuity — how the agent uses workspace files across sessions

USER.md

Context about the human user:

Name, pronouns, timezone
Working style preferences
Notes about current projects
What the user cares about

TOOLS.md

Tool-specific notes and instructions for the agent. This is where you can add domain-specific guidance about how to use tools.

Editing personas

During cluster creation

The persona editor is built into the cluster creation form. Each agent tab has IDENTITY, SOUL, USER, and TOOLS sub-tabs.

After creation

Go to the agent’s config page (click the agent name → Configure). Edit the persona and save. Changes are saved to the database — reprovision the cluster to apply them to the live runtime.

Persona versioning

Tulsk tracks persona versions. When you save a new persona, it creates a version snapshot. The cluster settings page shows whether the running version matches the saved version — if they differ, you need to reprovision.