What changed
The official guide distinguishes attaching a document's full text to a chat thread from embedding it for retrieval across a workspace. It applies to AnythingLLM version 1.8.5 and above; that version's changelog describes the file-chat overhaul. Attachments stay within their thread, while embedded documents become available across workspace threads. These sources establish the workflow and version context, but the opened pages do not establish a calendar launch date. [1] [4]
Why it matters
Editorial analysis: this could help a coordinator locate a policy detail before drafting a briefing, rather than treating an AI answer as the policy itself. Mintplex Labs' repository lists PDF, TXT and DOCX support, drag-and-drop uploads and source citations. Those documented features provide a starting point for checking answers, not proof of accuracy or measured time savings. [2]
Who it affects
Operations staff, project coordinators, consultants and team leads who have permission to use the files. Start with one narrow workplace question and a small approved document set. This is a briefing aid, not an authority for HR, legal or compliance decisions.
What you can do with this
Download the desktop app through the official site, which offers macOS, Windows and Linux versions and model selection during setup. [3] For a document-only pilot, choose a local language model and confirm that the embedding model and vector storage are also local. Before adding workplace files, explicitly turn off web search and avoid agent tools that contact external services. The official product site says web search is on from the first chat. Check the controls in the version you use and review the app's Privacy settings; a local chat model alone does not establish a no-network configuration. [2] [3] Create one workspace and drag an approved travel-policy file into its chat. Attach full text when it fits the model's context window; use Embed for larger files or reuse across threads. [1] Ask: 'What reimbursement rules apply to this trip? Identify supporting passages and say when the documents do not answer.' Compare the response and any citations with the original policy before writing the briefing.
What we don’t know
Retrieval supplies selected passages rather than every document in full, so relevant information can be missed. The guide warns that Continue Anyway with an oversized attachment prunes data and can produce inaccurate behavior. Embedded documents are available throughout their workspace and, in multi-user mode, to everyone with workspace access; the repository identifies multi-user support as Docker-only, not a desktop feature. [1] [2] Mintplex Labs describes local processing and telemetry opt-out in its desktop privacy policy, effective July 14, 2025. Its repository also documents anonymous usage telemetry and outbound connections for external providers, model downloads and other resources. These are company statements, not an independent privacy audit. [2] [5] This build was checked against documentation, not installed or tested; citations do not guarantee a correct answer.
Correction · Nov 19, 2026
Updated the source-status description and clarified the document-only setup to explicitly turn off web search, following the official product site's company-described default. The workflow remains documentation-based and untested.
SHOWING OUR WORK
Go straight to the source.
Facts are grounded in the references below. Practical implications are editorial analysis unless explicitly attributed to a source.
01See claim-level references
The repository identifies AnythingLLM as MIT-licensed and documents a no-code document-chat application; this does not assert licensing rights to every third-party model or component. Source
The official guide applies to AnythingLLM version 1.8.5 and above and distinguishes full-text chat attachments from embedded workspace documents. Source
Chat attachments are thread-scoped; embedding a document makes it available across workspace threads and, in multi-user mode, to every user with workspace access. Source
The v1.8.5 desktop changelog describes a file-chat overhaul that uses full file content when the context window permits and offers embedding for oversized files; it displays no explicit calendar publication date in the opened text. Source
The repository lists PDF, TXT and DOCX support, drag-and-drop uploads and source citations; these are documented project features, not independently tested accuracy results. Source
The company product site offers a desktop app for macOS, Windows and Linux and describes choosing a model during setup. Source
The repository documents local and external language-model, embedding-model and vector-database options; local chat-model selection alone does not establish a no-network setup. Source
The document guide instructs users to drag files into chat and offers Embed when full text exceeds the model context window. Source
The document guide describes retrieval as supplying limited semantically relevant passages and explains that filtering can exclude a passage containing the answer. Source
The guide warns that Continue Anyway with an oversized attachment prunes information and can cause inaccurate model behavior. Source
The repository labels multi-user instance support and permissioning as Docker-only. Source
Mintplex Labs' desktop privacy policy states an effective date of July 14, 2025, describes local processing and provides an in-app telemetry opt-out. The effective date is not treated as a verified publication date. Source
The repository documents anonymous usage telemetry, disabling it through the in-app Privacy control, and outbound connections for external tools or providers, model downloads and other resources even when telemetry is disabled. Source
The official product site's September 2026 blog listing says web search is on from the first chat. This is a company statement; the installed-version behavior, exact setting path and rollout day were not tested or established. Source