Magus is one workbench where every instrument shares a single context: autonomous agents, a knowledge brain, a code sandbox, your files and prompt library, and the integrations that reach the tools you already run.
Build named specialists, @mention them, and let them work autonomously, even waking themselves to finish long jobs.
Reason, call a tool, observe the result, and repeat, streamed live.
Spawn sub-agents, or convene a council of frontier models and consolidate their answers.
Heartbeat wake-ups process the inbox and long tasks.
A markdown knowledge base with [[wikilinks]] and a cross-resource graph that searches everything you've written, uploaded, or said.
Author links in-text; the graph indexes them on save.
pgvector over pages, files, drafts, and memories.
Local, user, and agent scopes the agent can recall.
Two panes at once: a live conversation on one side, your work on the other, so you can draft, read, and annotate without losing your place. 15+ models auto-routed, plus image and video.
Keep a chat live on one side; draft, read a brain page, or open a file on the other.
Pick the model, or let the router choose by task.
Write documents side by side with the agent, view and annotate PDFs, and export when you're done.
Beyond the headline instruments: a code sandbox, your files and prompt library, real-time collaboration, and the integrations that reach your stack, all sharing the same context.
Open a page beside a chat, hand a file to an agent, watch a sandbox run, all without losing your place.
A small Swiss team, building with the community. The core is open source on GitHub and the conversation lives on Discord. Come shape it with us.