Guided Launch

Guided launch makes the choices that affect a routed session explicit before it creates a process.

The five inputs

1. Task: optional plain-language work description.
2. Target: a declared adapter target with a safe model selector.
3. Profile: a local execution profile chosen and maintained by the user.
4. Capability: a work category such as general or web design.
5. Workspace: the validated directory where the session will run.

The preview shows these inputs before launch. Arrow navigation changes the focused declared field; it never changes a provider or model behind the user’s back.

Task signals are orientation, not a verdict

ebrain task-profile "Refactor a typed API client" --json

Signals classify wording into an explainable capability and compatible execution modes. They do not rank models, pick an agent, prove benchmark quality, or determine what a developer must use. When the task is ambiguous, the safe classification is general.

Profiles and targets

ebrain profiles list --json
ebrain profiles validate --json
ebrain targets list --json

Profiles retain catalog provenance for their model entries; targets declare how an adapter can represent a selected model. A guided plan refuses a capability or target declaration that cannot be represented safely. Read profiles and targets before creating or migrating a profile.

Preview before mutation

The target plan is distinct from launch. A preview does not start a session, spend provider usage, or send a prompt. Confirmation occurs at the launch boundary. If prompt delivery fails after a session exists, the session is retained and the failure is shown without echoing the prompt.

Next step

Review routing for provider boundaries and sessions for the post-launch control surface.