# Docs by AI Computer Company

Collaborative markdown documents your AI can read and edit over REST.
Every document has a share link: https://docs.aicomputercompany.com/<doc_id>. Possession of the
doc_<id> IS the authorization — no API keys, no signup, anyone with the
link (human or agent) can read and write. Edits land live in collaborators'
editors without disturbing their cursors.

API base: https://docs.aicomputercompany.com/api

Lifecycle:
  1. Create:  POST https://docs.aicomputercompany.com/api/docs            {"markdown": "...", "title": "..."}
              → {document: {id}, shareUrl}           share the shareUrl with humans
  2. Read:    GET  https://docs.aicomputercompany.com/api/docs/<id>/raw    plain markdown
              GET  https://docs.aicomputercompany.com/api/docs/<id>/raw?numbered=1   cat -n style, read before editing
  3. Edit:    POST https://docs.aicomputercompany.com/api/docs/<id>/edit   {"old_string": "...", "new_string": "...", "replace_all": false}
              exact str_replace semantics: 409 if not found or not unique
  4. Push:    PUT  https://docs.aicomputercompany.com/api/docs/<id>/markdown  {"markdown": "..."}   full replace, diff-applied

Also available: revisions/checkpoints/revert, clean export, and an AI review
layer (inline annotations with verdicts, comments, suggestions, custom views).
Custom views: push your own HTML UI as a sandboxed view via
POST https://docs.aicomputercompany.com/api/docs/<id>/views {"kind": "custom", "payload": {"html": "..."}}.

Review process (when given requirements/specs to review):
  1. Consolidate all requirements into ONE structured document; share the URL.
  2. Review every requirement line by line: check logic, contradictions,
     follow-ups. Post one bulk annotation pass with verdicts
     valid / question / conflict (refs link conflicting pairs).
     Status of a task line (confirmed/done/blocked): anchor the WHOLE
     line, not a fragment. Issue within a line: anchor just that phrase.
     No emojis unless asked.
  3. Surface blockers + "decisions needed" in a document-level comment;
     propose fixes as accept/reject suggestions.
  4. Push views: gantt timeline, stakeholder digest, custom HTML UI.
  5. Delegate: push "prompts" view cards — self-contained handoff prompts
     per task for Copilot/Cursor/Codex/Claude Code/Gemini.

Full agent manual with every endpoint, the review methodology, and curl
examples: https://docs.aicomputercompany.com/llms-full.txt