Sources you choose
Everything you may see through Workspace Search, chosen items, or PDFs, text files and pasted text for one conversation.
Answers from your work, with the sources shown
Ask questions about your own workspace and files, get answers with citations, and turn them into drafts.
Conversations are private to the person (a team member's email, or the domain's owner for the owner and deployment passcodes); owners and the admin get a 404 for anyone else's conversations, files and drafts. Each question picks its sources: everything the person may see (a Workspace Search query as that person, the best 8 with full text), up to six chosen items re-read on every question, or PDFs up to 10 MB, text and Markdown files and pasted text for that conversation (R2). Claude (claude-opus-5-5) answers from search-result and document blocks with citations turned on, streamed to the page; every citation opens the item or the file with the quoted passage, and when the sources do not cover the question the answer says 'I could not find this in your sources.' (with no sources found, Claude is not called at all). Quick actions summarise, list decisions and action items, compare two items, draft a reply and show what changed this week. Drafts (email, document, note) come from an answer or are written by hand; they are edited and copied, opened in the person's email app through mailto (nothing is sent), or saved to Writer (POST /api/v1/documents) and Notes (POST /api/v1/notes) as private items with the sources listed. 30 questions or AI drafts an hour per person, and each person's monthly questions and tokens shown to owners. Without an Anthropic key the app says what it needs, and drafts still work by hand. No public pages, no feed of its own and no webhooks.
What Assistant does for you.
Everything you may see through Workspace Search, chosen items, or PDFs, text files and pasted text for one conversation.
Each answer cites its sources; when they do not cover the question, it says it could not find it instead of guessing.
Email, document and note drafts you edit, copy, open in your email app, or save to Writer and Notes.
A suggested path for Assistant. What you can do depends on your access in the app.
Everything I can see, chosen items, or files for this conversation.
Type a question, or use a quick action such as Summarise or Decisions and action items.
Turn the answer into an email, a document or a note, and edit it before you use it.
Live host, inventory, governance, health checks, verification, dependencies and history.
Conversations are private to the person (a team member's email, or the domain's owner for the owner and deployment passcodes); owners and the admin get a 404 for anyone else's conversations, files and drafts. Each question picks its sources: everything the person may see (a Workspace Search query as that person, the best 8 with full text), up to six chosen items re-read on every question, or PDFs up to 10 MB, text and Markdown files and pasted text for that conversation (R2). Claude (claude-opus-5-5) answers from search-result and document blocks with citations turned on, streamed to the page; every citation opens the item or the file with the quoted passage, and when the sources do not cover the question the answer says 'I could not find this in your sources.' (with no sources found, Claude is not called at all). Quick actions summarise, list decisions and action items, compare two items, draft a reply and show what changed this week. Drafts (email, document, note) come from an answer or are written by hand; they are edited and copied, opened in the person's email app through mailto (nothing is sent), or saved to Writer (POST /api/v1/documents) and Notes (POST /api/v1/notes) as private items with the sources listed. 30 questions or AI drafts an hour per person, and each person's monthly questions and tokens shown to owners. Without an Anthropic key the app says what it needs, and drafts still work by hand. No public pages, no feed of its own and no webhooks.
Finding items belongs to Workspace Search; source notebooks and research briefs belong to Research; the website chatbot for visitors belongs to Chatbot.
5 of 18 required controls are configured. Catalog deployment and inventory maturity remain independent until every required control is present.
assistants.appneural.com
No remediation required.
No remediation required.
No remediation required.
No remediation required.
Configure or repair: Repository exists.
Configure or repair: Business owner exists.
Configure or repair: Product owner exists.
Configure or repair: Technical owner exists.
Configure or repair: Maintaining team exists.
Configure or repair: Documentation exists.
Configure or repair: Architecture documentation exists.
Configure or repair: Security review is current.
No remediation required.
Configure or repair: Dependencies exist.
Configure or repair: Consumers exist.
No remediation required.
Configure or repair: SLA or SLO exists.
Configure or repair: On-call exists.
Configure or repair: Release exists.
No remediation required.
Configure or repair: Infrastructure drift is clear.
No remediation required.
No remediation required.
No remediation required.
No remediation required.
Configure or repair: Repository exists.
Configure or repair: Business owner exists.
Configure or repair: Product owner exists.
Configure or repair: Technical owner exists.
Configure or repair: Maintaining team exists.
AI outputs are saved as drafts with evidence references. They never replace verified catalog fields automatically.