Model, session, and profile UX

9 min readFor operatorsUpdated 2026-05-22

Craik v0.10.0 adds operator-facing commands for model selection, persistent session inspection, and local profile/persona boundaries. These commands are designed for shell discovery and automation-friendly JSON output.

Model selection

Set the active model with a provider-qualified reference:

craik model set openai/gpt-5
craik model status

The selected model is stored under $CRAIK_HOME/config/model-settings.json and appears in craik status and the shell startup card.

Aliases and fallbacks are managed explicitly:

craik model alias list
craik model alias add fast openai/gpt-5-mini
craik model fallback add anthropic/claude-sonnet
craik model fallback list

Unknown token and cost data is reported as unknown until a provider adapter supplies concrete usage metadata.

Sessions

Persistent agent session records are read from the local runtime store:

craik session list
craik session show agent_session_docs
craik session resume agent_session_docs
craik session export agent_session_docs

Exports are redacted by default. prune and delete require --yes. Deletion marks a session stopped rather than removing the record, which preserves continuity and audit history.

Session and usage commands require an active operator session because they inspect persisted runtime activity, receipts, and handoff links.

Profiles and personas

Profiles give operators a local boundary for provider config, model selection, sessions, skills, memory, and gateway state.

craik profile create release --description "Release work"
craik profile use release
craik profile list
craik profile export

The current active profile is visible in craik status and the shell startup card. Profile exports omit secrets by default.

Usage and insights

Use craik usage or craik insights to summarize known runtime activity:

craik usage
craik insights

The summary names unknown values explicitly instead of silently omitting them. This keeps early local use honest while leaving a stable place for provider adapters to attach token, cost, failure, approval, and skill impact metrics.