Install and prove the computer. uv sync, then uv run design-agent sandbox build (Docker; a cold build downloads about 2 GB and takes several minutes) and uv run design-agent demo: a scripted run renders the sample brand's launch card in three sizes and a 4-second teaser with no keys. If it fails, uv run design-agent doctor says what is missing.
Model. uv run design-agent config set DESIGN_AGENT_MODEL=provider:model and the provider's key (ANTHROPIC_API_KEY=…, or LANGSMITH_GATEWAY_API_KEY for a langsmith: model; typed by the person, or set in .env by hand). Defaults and other roles: docs/models.md. Re-run doctor, then uv run design-agent test model: doctor only checks that a key is set, the test makes one cheap call with it.
Media keys (optional). OPENAI_API_KEY (image generation), ELEVENLABS_API_KEY (voice, music, sound effects), GROQ_API_KEY (fast transcription). Each tool says plainly when its key is missing.
Brand. Follow the brand-onboarding skill to build workspace/brand/ from what the person has (a brand guide, a website, files, or a Paper file). Until then the sample brand is used.
Paper (optional). PAPER_MODE=local when Paper Desktop runs on this machine; bridged for a deployed agent (see docs/partners/paper.md). off is fine: boards render in the sandbox.
Try it. uv run design-agent ask "Make a LinkedIn card announcing …". Finals land in workspace/out/<thread>/, the last line estimates what the run cost, and --thread <id> continues a conversation.
Choose where it runs. Managed Deep Agents (managed runtime, sandboxes, Slack app and schedules; one command) or self-hosted (Docker Compose with Postgres and their own Slack app). The design-agent-deploy skill takes it from there, including verifying with a real request.
What next. Teach it the deliverable the team asks for every week (design-agent-customize), and measure changes with design-agent-evals.

design-agent-onboarding

Set up Design Agent