Skip to main content
This page covers the mental model behind Whim. Understanding these concepts will help you get the most out of the platform.

​
Workspaces

A workspace is your project container in Whim. Each workspace is linked to one or more GitHub repositories and acts as the home for all tasks, configuration, and team collaboration on that project. When you create a workspace, Whim:
  1. Syncs your repository’s code into the cloud
  2. Sets up the default AI provider and configuration
  3. Creates a shared space where your team can launch and monitor tasks
Think of a workspace like a project in your IDE — it’s the context in which all your AI agents operate.
You can have multiple workspaces, each linked to a different repo. Switch between them from the workspace selector in the top navigation.

​
Tasks

A task is a single AI agent session running in its own isolated cloud container. When you create a task, Whim:
  1. Spins up a sandboxed Ubuntu container on Fly.io
  2. Clones your repo and checks out a fresh git branch
  3. Starts the AI coding agent (Claude Code, Codex, or OpenRouter)
  4. Streams the terminal output to your browser in real time
Each task has its own:
  • Container — isolated environment with shell access
  • Git branch — automatic branch creation, no conflicts with other tasks
  • Terminal — real-time streaming of agent output
  • Conversation history — full record of prompts, reasoning, and actions

​
Task lifecycle

Tasks move through these states:
StateDescription
Active / RunningThe AI agent is actively working — executing commands, editing files, reasoning.
WaitingThe container is running but the agent needs input or is idle.
SleepingThe container is suspended to save compute. Auto-sleep kicks in after a configurable idle period. Tasks resume instantly when you interact with them.
CompletedThe task is done. The branch and results are preserved.
ArchivedThe task is archived and its container is suspended.

​
Simple vs Orchestrator mode

Whim supports two task modes:
  • Simple mode — One agent, one task, one prompt. The default for most work.
  • Orchestrator mode — A coordinator agent that breaks down complex work and spawns subtasks. The orchestrator delegates to child agents that work in parallel, then aggregates results. Useful for large features, multi-file refactors, or work that benefits from divide-and-conquer.

​
Todos

A todo is a planned work item in your backlog. Unlike tasks, todos don’t run immediately — they’re placeholders for work you want to do later. You can:
  • Create todos with a title, description, and priority
  • Assign todos to team members
  • Drag and drop to reorder your backlog
  • Launch a todo into a running task when you’re ready
Todos are useful for tracking planned work, triaging bugs, or building a queue of prompts for your AI agents.

​
AI Providers

Whim supports three AI providers, each running their native CLI:

Claude

Anthropic’s Claude Code CLI. Supports Claude Opus, Sonnet, and Haiku models. Can use Whim’s built-in compute or your own Claude Pro/Team subscription.

Codex

OpenAI’s Codex CLI. Requires your own OpenAI API key. Supports configurable reasoning effort levels.

OpenRouter

Access 10+ models from multiple providers (GPT, Gemini, Grok, DeepSeek, and more) through a single API key.
Each provider has its own permission modes that control how much autonomy the agent has:
  • Default — Prompts for approval on first use of each tool
  • Accept Edits — Automatically accepts file edits (Claude)
  • Bypass Permissions — Skips all permission prompts (Claude)
  • Plan Mode — Can analyze code but not modify files (Claude)
You can configure providers per workspace and switch between them on a per-task basis.

​
Branch isolation

Every task in Whim gets its own git branch, created automatically when the task launches. This means:
  • No conflicts — Multiple tasks can work on the same repo simultaneously without stepping on each other
  • Clean PRs — Each task’s changes are isolated on their own branch, ready for code review
  • Safe experimentation — Try multiple approaches in parallel; keep the one that works
The branch naming follows the pattern whim-<workspace>-<task-id>. When a task creates a pull request, it’s linked directly from the task view. You can also set a source branch — the base branch the task should branch from (defaults to main).

​
Compute Units

Whim measures usage in Compute Units (CUs). CUs are a combined measure of:
  • Container runtime — How long your task containers run
  • AI token consumption — Input and output tokens used by the AI models
Your plan includes a weekly CU allocation. When you’ve used your allocation, tasks pause until the next cycle or you upgrade your plan.
CUs are designed to be simple and predictable. You don’t need to worry about separate bills for compute and tokens — it’s all bundled into one number.

​
Plans overview

PlanCUsBest for
TrialLimited weekly allocationTrying Whim out, no credit card required
ProHigher weekly allocation (scalable)Individual developers with regular AI agent usage
TeamHigher weekly allocation (scalable)Teams with collaboration, shared workspaces
All plans include the full feature set. Annual billing is available at a discount.

Creating Tasks

Learn how to create and manage tasks in detail.

Onboarding Walkthrough

Step-by-step guide through the setup wizard.