reelyai

Local agent CLI for creating and driving ReelyAI web workflows.

The default target is https://reelyai.app, and you can point it at a local server with REELYAI_AGENT_BASE_URL or --base-url.

npm install -g ./packages/reelyai-cli
reelyai workflow "一个失眠导演在午夜便利店遇见未来的自己" --duration 60

From npm:

npm install -g reelyai
reelyai skill install --agent all

The package bundles the reelyai-cli skill. reelyai skill install --agent all copies it into local Codex, Claude, Cursor, and generic .agents skill folders.

For deployments protected by REELYAI_ACCESS_TOKEN:

reelyai configure --base-url https://reelyai.app --access-token "$REELYAI_ACCESS_TOKEN"

For browser-scoped Agent Plan generation:

reelyai configure --agent-plan-token "$ARK_AGENT_PLAN_KEY"

Main commands:

• reelyai workflow "<idea>": create a session, generate script, and generate storyboard/workflow.
• reelyai status --session latest --deep --json: inspect shots, renders, errors, stitch state, and download URL.
• reelyai render --session latest --stitch --progress: generate missing shots and stitch the final video with visible progress.
• reelyai render --session latest --stitch --stitch-partial --repair-policy safe-retry --max-attempts 2: retry policy failures with a safer prompt and stitch ready shots.
• reelyai download --session latest --output ./final.mp4: save the final video locally without custom fetch scripts.
• reelyai handoff --session latest --open: create a one-time handoff link so a normal browser can claim and edit a CLI-owned session.
• reelyai open --session latest: open the session URL (#/s/<sessionId>).

Recommended agent flow:

reelyai workflow "<idea>" --duration 60 --json
reelyai status --session latest --deep --json
reelyai render --session latest --stitch --progress --json

workflow --json includes both webUrl and handoffUrl. Because CLI and browser identities are isolated by reelyai_user_id cookies, return handoffUrl when a human needs to see and continue editing the AI-created workflow in a normal browser. The raw webUrl is still useful inside the CLI cookie scope.

Use --jsonl for long fully automated runs; it emits session_created, shot_submitted, task_id, poll_status, retrying, and stitch_ready events as newline-delimited JSON.

The CLI stores only local configuration and cookies under ~/.reelyai/config.json. The ReelyAI app remains the source of truth for sessions, shots, prompts, renders, stitch jobs, and final videos.