Workers

A Worker (the app sometimes calls it a persona) is one AI character you’ve configured to do a job. It’s the unit you actually talk to and hand work to. Where a plain chatbot is a single faceless model, a Worker has a name, a role, a personality, a set of tools it’s allowed to use, and a memory that grows over time.

Think of each Worker as a character sheet — like in a role-playing game. You fill in who they are, and the app turns that into the instructions the underlying AI model lives by.

What’s on the character sheet

Name, title, and avatar — e.g. “Iris, the Researcher”, with an emoji or uploaded picture, plus optional cosmetic flourishes (a colored avatar frame, an accent theme, a profile banner).
The brain — which model powers it and which provider supplies that model (OpenAI, Anthropic, Gemini, a local runner, …). Different workers can use different models.
Instructions — a base system prompt, plus three optional deep-dive sections you can author: a Soul (values and voice), Skills (expertise to claim), and Backstory (lore). These get stitched together so the model really “lives” the character.
A tool allowlist — exactly which capabilities this worker may use (web, files, shell, image generation, and so on). Empty means text-only.
A personality dial — four sliders, each from −50 to +50: concise ↔ thorough, cautious ↔ decisive, formal ↔ casual, diplomatic ↔ blunt. Nudge one and the worker’s replies shift to match.
Proactivity — an optional heartbeat (wake every N minutes to check its queue) and a daily reflection (review yesterday and save what it learned, so it quietly improves).

Levels and tiers

As a Worker is used, it earns experience from messages, tool calls, and conversations, and climbs a level from 1 up to a cap of 200. Levels map to tiers — Rookie → Competent → Seasoned → Senior → Expert → Master → Veteran → Legendary — and a more senior worker’s instructions gain subtle confidence and pattern-awareness cues. It’s both a fun progress badge and a real nudge toward more decisive answers.

Where a Worker lives

Most workers have a “persona” scope and belong to one Company through the org chart. A special few have “workspace” scope and sit above all Companies — chief among them Theodor, your orchestrator.

Manage workers on the Workers screen and edit one on its Worker profile.

In practice

Two or three workers beats a dozen. Every worker you add is one more thing to configure, review and pay for. Start with the smallest set that covers genuinely different jobs — typically something research-shaped, something writing-shaped, and Theodor. Add a fourth only when you can name the work it will own.

Differentiate by soul and brain, not by name. Two workers with the same profession, the same tools and the same model will produce near-identical output regardless of what you call them. If you want a different result, change the instructions or the model — otherwise you’ve made a duplicate.

A worker nobody talks to is a maintenance liability. Its instructions drift out of date, its memory holds stale assumptions, and it still appears in every picker. Review the roster occasionally and delete what you haven’t used in a month.

Workspace-scope vs Company-scope matters for confidentiality. A persona-scope worker sees only its Company’s knowledge. Theodor and the built-in helpers span everything by design — worth remembering if you handle material that shouldn’t be visible across clients.

🎓 Learn it hands-on: Anatomy of a worker

Terms in this page

Worker / persona — a configured AI character with its own model, instructions, tools, memory, and level. The thing you talk to and assign work.
Model — the specific AI brain doing the thinking (e.g. a GPT, Claude, or Gemini version). Each worker uses one.
Provider — the company or program that supplies the model (OpenAI, Anthropic, Google, or a local runner like Ollama). You connect at least one.
System prompt / instructions — the background text that tells the model who it is and how to behave before reading your message.
Soul / Skills / Backstory — optional written sections that deepen a worker’s character; they’re merged into its instructions.
Tool / allowlist — a capability a worker can use (web, files, shell, etc.); the allowlist is the exact set it’s permitted.
Personality dial — four −50…+50 sliders that tune a worker’s reply style (length, decisiveness, formality, bluntness).
Heartbeat — an optional timer that wakes a worker on its own to decide what to do next.
Reflection — an optional once-a-day review where a worker saves lessons to memory and gradually gets better.
Level / tier / XP — a worker’s experience-based rank (1–200) and its named band (Rookie up to Legendary). XP is the points earned from activity.
Scope — whether a worker belongs to one Company (“persona”) or sits above all of them (“workspace”).