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
"""MDA memory: a team layer every caller shares, and a personal layer per person.
`/memories/agent/` holds team rules (anyone who mentions the agent can add to it, so it is never a
place for private data). `/memories/user/` belongs to the verified person who asked; MDA mounts it
only for a run with a person behind it.
Declared here, the personal layer mounts where MDA's defaults put it: the person's Slack DM and
Studio. People ask for work in channel threads, so `agent.py` selects both layers for every Slack
run (`slack_run`) and their preferences follow them there. The cost: their hot `AGENTS.md` is in
context in a shared thread, which is why `instructions.md` says never to repeat it there.
"""
from managed_deepagents import ManagedServerRuntime, MemoryLayer, define_memory
memory = define_memory(agent=MemoryLayer(), user=MemoryLayer())
def slack_run(runtime: ManagedServerRuntime) -> bool:
"""A run a Slack event started (a mention, a DM, a thread reply or an approval click).
MDA sets the runtime's channel only for a channel run, from its own routing, never from the
context a caller sends; inspection, Studio, API and scheduled runs have none.
"""
return runtime.channel is not None and runtime.channel.provider == "slack"