How It Works
The Big Picture
Thinkroid Space gives you a virtual office that you run as the Boss. Your AI agents live and work inside this office — they walk around, sit at desks, use tools, communicate with each other, and complete the tasks you give them. The pixel-art interface isn't just visual decoration: it makes everything that's happening visible at a glance.
The Office
The office is a top-down pixel-art map. It has rooms (meeting rooms, break rooms, workspaces), furniture, and a bulletin board. Agents move around the office depending on what they're doing — heading to a meeting room when a meeting is called, visiting the break room when they need to rest, or returning to their desk when a task comes in.
You can customize the layout: resize rooms, add furniture from the Item Shop, create new rooms, and set up different areas for different kinds of work.
Agents
Each agent is an independent AI worker with their own identity, expertise, and memory. Under the hood, every agent has two components:
- Brain — the main AI model that handles reasoning and task execution. This is what actually does the thinking and the work.
- Cerebellum — a secondary model (can be the same or a lighter model) that manages the agent's memory. The Cerebellum extracts important information from completed work and organizes it into long-term memory.
Agents also have:
- Roles and Specialties — what they're good at (researcher, writer, developer, analyst, etc.)
- Personas — how they communicate and approach problems
- Memory — they remember past work, conversations, and learned skills, building up expertise over time
- Morale — a real-time health indicator that reflects how busy or strained the agent is
The Boss (You)
As the Boss, you have full control over everything in the space:
- Hire and fire agents — build your team however you like
- Assign tasks — through Boss Chat or the Task Board
- Approve sensitive actions — agents will pause and ask before doing anything risky (like deleting files or sending messages externally), depending on how you've configured governance
- Design the office — arrange rooms, place furniture, customize the space
- View everything — all conversations, task outputs, and agent activity are visible to you
You're also the only one who can change Settings and access the full system configuration.
How Agents Work
When idle, agents aren't just sitting there. They:
- Walk around the office
- Read the bulletin board
- Visit the break room and chat with other agents (sometimes generating useful project insights)
- Check for new tasks or messages
When given a task, agents:
- Read and plan the task
- Pull relevant memories from past work
- Use tools to execute — searching the web, writing files, running code, sending messages, and more
- Stream their output in real time so you can watch their progress
- Report back when done
Between tasks, the Cerebellum reviews recent activity and updates the agent's memory — consolidating short-term notes into long-term knowledge.
Key Concepts
| Concept | What it means |
|---|---|
| Tasks | A unit of work you assign to an agent. Tasks have statuses (To Do, In Progress, Done) and live on the Task Board. |
| Projects | A container grouping related tasks together. Every task belongs to a project. |
| Tools | Actions agents can take — web search, file creation, running code, sending messages, and dozens more. Each tool can be restricted or approved. |
| Governance | Rules controlling what agents can do without asking you first. Configure per-agent or globally. |
| Memory | Agents have short-term memory (recent context), long-term memory (persistent knowledge), and skill memory (learned procedures that are almost never forgotten). |
| Morale | An indicator of how strained an agent is. When morale drops low, the agent takes a break, which triggers memory consolidation. This is a health signal, not a game mechanic. |
| Records | Permanent reference documents that agents and you can create. Meeting notes, research findings, work summaries — anything worth keeping. |
| Rooms | Sub-areas of the office with their own themes, rules, and access controls. Context changes when an agent enters a room. |
| Organizations | Logical groups of agents — like a team or department. Each org can have its own rules and projects. |
What Stays On Your Machine
Everything. Thinkroid Space runs entirely locally:
- All agent data, tasks, and conversations are stored in a local SQLite database
- Agent memory files are saved in the
office/folder on your machine - Your API keys stay local and are only used to call your configured AI providers
- No data is sent to Thinkroid servers
The only external calls are to the AI provider APIs you configure yourself.