Muse MuseConcept

Independent concept built on Meta's Model API. Not an official Meta product.

Brief Desk

Status: Live.

  1. Your muse

    Build Brief Desk: a page where I paste a long made-up policy document and ask questions, with streamed answers, tool calling and retries, plus a triage copilot with side chats off the main conversation and validated find-and-replace edits on the document. Use the model API patterns and short agent patterns recipes. Made-up data only.

  2. Done

    Checks passed

  3. The plan: Brief Desk

    Ask questions of a draft policy with streamed answers and safe edits.

    1. Paste a made-up policy document into the desk
    2. Ask a question in the main conversation
    3. Read the streamed answer with quoted sections
    4. Open a side chat for follow ups or triage
    See the full plan

    Who it's for

    People drafting and reviewing made-up policy documents.

    How it works

    1. Paste a made-up policy document into the desk.
    2. Ask a question in the main conversation.
    3. Read the streamed answer with quoted sections.
    4. Open a side chat for follow ups or triage.
    5. Apply a validated find and replace edit and copy the result.

    Screens

    • Desk. Paste the document, chat in main thread, and see streamed answers.
    • Side chat. Run triage and follow up chats off the main conversation.
    • Edit review. Check validated find and replace matches before applying.

    What goes in

    • Text: Long made-up policy document pasted by the user, up to about 200k words.
    • Text: Question about the document typed in the main chat or a side chat.

    What comes out

    • Text: Streamed answer with section quotes from the pasted document.
    • Text: Validated find and replace preview showing matches and changed text.

    Models

    ModelUsed forWhy
    Muse WonderAnswer questions over a long pasted document and draft editsReads long text context, streams answers, does tool calling and retries, and returns JSON for validated edits.

    Limits

    • Made-up data only. Not legal advice and not for real policies.
    • Answers are generated and can miss or misquote sections. Check quotes before use.
    • Document stays in the browser unless exported. No accounts or shared storage.
    • Very long documents may be truncated to fit the model context.
    • Muse Code workspaces are not included.

    From your inspo

    No references were given. The note asks for a Brief Desk page to paste a long made-up policy, ask questions, and edit with copilot help.

    Takes

    • Question answering over a pasted policy document
    • Streamed answers with retries and tool calling
    • Triage copilot with side chats and validated find and replace edits

    Cost per run: about $0.18 (est.). One run uses about 40k input tokens at $4 per M for $0.16 plus about 1000 output tokens at $20 per M for $0.02, total $0.18.

  4. Recorded demo
  5. Done. Brief Desk is done. npm run verify passes all steps on the final tree.

    32 m 59 s · Input 13M · Output 60k · $0.06 est.