Tasks vs Docs vs Memory
BEAM has three persistence surfaces, and they are not interchangeable. Work you're going to do, things you've written down, and facts BEAM has learned about you each live in a different place with a different lifecycle. Knowing which surface something belongs on is most of the skill of using BEAM well.
Tasks: bounded units of work
A task is something that can be done — it has a title, maybe a due date, a priority, a list, and eventually a checkmark. Tasks live inside lists inside projects, can nest one level (subtasks), and can block each other (dep, block). The lifecycle is explicit: created → active → done → cleared or trashed → restored or purged. Nothing skips a step, and nothing is unrecoverable until you purge.
The tell that something is a task: if it can never be finished, it isn't one. "Rehearse the closing" is a task. "Thoughts on how endings work" is not — that's a doc.
Docs: long-form content
Docs are where prose lives: notes you write in the editor, files you drag in (PDF, DOCX, and images render faithfully with the original preserved), research reports deep saves, answers you keep with --save or shelf this. Docs sit in projects, optionally in folders, and link to tasks in both directions — a task can carry its supporting material, a doc can point at the work it implies.
Docs are also what the AI grounds on. When you ask about your work, BEAM searches doc content — so a decision written into a doc is a decision the AI can find, quote, and build on later. Text trapped in your head or another app is invisible to it.
Memory: what BEAM knows about you
Memory is the smallest and strangest surface: facts about you and how you work, held with explicit uncertainty. "Prefers morning rehearsals." "Tends to overrun on Q&A." Each memory carries a confidence zone — ● established, ◐ likely, ○ tentative, ◌ untested — and confidence moves with evidence: confirmations strengthen it, contradictions weaken it, and you can adjudicate directly with memory test <ref> confirm|reject.
Memories accrue two ways: you add them (memory add), or BEAM extracts them from what you say in AI conversations — never from what the AI said, which keeps the record grounded in your own words. The AI reads memory when it advises you, which is why brief gets more specific after a few weeks: it isn't guessing at your patterns anymore.
How the AI sees the three
When you ask something, the AI looks — it searches tasks for commitments, docs for substance, and memory for context about you. Each surface answers a different kind of question: tasks say what's true about your obligations, docs say what's true about your thinking, memory says what's usually true about you. If the answer to a question should survive the question, put it in a doc; if it should get done, make it a task; if it's about you rather than the work, it belongs in memory — and BEAM has probably already noticed.