Scenarios: what do you want to do?
Updated August 10, 2026
The rest of the docs are organized by feature: here is the worktree system, here is the CLI, here is what the phone app does. That’s the right shape when you know what a thing is called. It’s the wrong shape when you arrive with an intent.
So this section is organized the other way. Pick the sentence that matches what you want, follow the steps, and each page hands you back to the feature reference when you want the whole story.
Every scenario is short on purpose. It’s a path through features, not a second copy of them.
working with models
- Run your coding agent on your own hardware. Point Claude Code or Codex at a model server you run, and keep the prompts on your machine.
- Cut model costs with hosted open models. Run Claude Code against a managed open model without handing it your Anthropic key.
- Teach crystl your models once. Map small, standard, and large to your own model ids so heroes and quests land on the right one.
running work
- Fan out cheap workers while your main agent leads. One session becomes a manager and runs a team of workers for you.
- Parallel feature work without merge conflicts. Two or more agents on one repo, each in its own worktree and branch.
- Run a multi-agent quest. Summon a party of agents that coordinate in a shared chat.
- Run agents on a schedule. The overnight triage or the morning inbox pass, launched without you.
- Work on a repo on your home server over SSH. Agents run on the remote box, approvals still land on your Mac.
staying in control
- Only get interrupted when an agent needs you. Tune approval modes and notifications so the quiet agents stay quiet.
- Supervise your agents from your phone. Watch, answer, and approve away from the keyboard.
- Recover context after closing or losing a session. Find what an agent did, and get the agent itself back.
- Review and merge a worker’s branch safely. Read the work first, then integrate it without forcing anything.
when you’d rather read the feature
Scenarios link to the reference pages rather than restating them. If you already know the feature name, start there instead:
- gems and shards, isolated sessions, open models, model sizes
- approval modes, notifications, history navigator
- fanout, agent scheduling, crystl quest, remote SSH sessions, mobile app
- CLI reference for every command a scenario uses