Open-source design agent: on-brand stills, decks, diagrams, charts and motion graphics from a chat thread, built on Deep Agents.
JavaScript agentagentsai-agentdeepagentsdesignhyperframeslanggraphmanaged-agentsmotion-graphicspaperslackvideo
CLI
uv run design-agent <command>; --help on any command for its options.
| Command | Does |
|---|---|
demo |
a scripted run through the real agent, sandbox and gates, no keys: a card in three sizes and a 4-second teaser |
doctor [--json] |
checks Python, packages, every model role, keys, the brand, the sandbox, MDA declarations, MCP and Paper, without printing secrets |
ask "<request>" [--thread ID] [--file PATH]… [--no-wait] |
one request with your model; stays open for background work; threads persist in workspace/state/ |
config show|set KEY=VALUE…|generate KEY… |
read and change .env without printing secrets |
brand validate|css|show|init |
check the active pack, regenerate tokens.css, show it, start one from the sample |
fonts verify |
brand font files present with their checksums; open fonts listed in fonts/LICENSES.md |
sandbox build|test|publish|use SNAPSHOT |
build the local image, run its probe, publish the LangSmith snapshot for MDA, point local runs at a snapshot |
skills verify|sync-vendor |
vendored HyperFrames, LangSmith and Deep Agents skills against skills-lock.json; re-vendor from upstream (--upstream for one, --check-only to compare pins) |
manifest |
what the workspace stager puts in each sandbox |
serve |
the self-hosted HTTP API (self-hosting) |
slack |
the self-hosted Slack app |
paper-bridge |
serve this machine's Paper Desktop to a remote agent behind a bearer token (Paper) |
eval list|run |
the eval set (evals); run spends money and asks first |
test model|sandbox|paper|slack|db |
real connectivity checks; model costs a cent or so |
There is no web setup console in v0.1: config, doctor and the coding-agent
onboarding skill cover setup.