The ButlerUseful technology. Clearly explained.
Home / AI setup guides
The Butler field guides

AnythingLLM: chat with your documents

Use the desktop app to organize a small document workspace, connect a model and verify your first answer before importing a whole library.

Documentation checked November 14, 2026. Installation paths have not all been hands-on tested. Follow the linked official requirements if they change.

Start with Desktop, not a server

For one person exploring document chat, start with AnythingLLM Desktop. A self-hosted server is a separate setup, useful when your requirements justify it. You need both an interface and a model: choosing a local model uses your hardware; choosing a cloud provider adds that provider's account, data handling and billing. Desktop overview. Documentation checked November 14, 2026; this is not a report of a hands-on install.

1. Download the right installer

Use the official downloads, not an advertisement or repackaged installer.

  • Windows: select the x64 or ARM installer for your hardware and install for Current User, not all users. Documentation targets Windows 11 and cautions that Enterprise/Server environments may differ. GPU dependencies may be installed during setup; missing acceleration can leave local models running more slowly on the CPU. Windows instructions.
  • Mac: choose Apple Silicon or Intel as appropriate. Open the DMG and drag the app to Applications. Mac instructions.
  • Linux: the official installer downloads the appropriate AppImage and offers desktop integration. Follow its instructions as your normal user, not root. Review any requested AppArmor rule before approving a system change. Linux instructions.

The developer suggests 16GB RAM and an eight-core CPU for a basic local-model setup, but the model is usually the limiting factor. Cloud inference reduces local hardware demands, not necessarily privacy or cost concerns. Hardware guidance.

2. Choose a model and one test document

  1. Complete onboarding. For local use, select the built-in model option or a working local provider such as Ollama. For a cloud provider, review its charges before supplying a key. See subscription versus API access.
  2. Create a workspace named My first test. Make a short text file containing three meeting dates or three product specifications you know are correct.
  3. Attach the file to the chat using the + control or drag-and-drop. Ask a precise question about one item.
  4. Compare the answer with the file. Then ask about a fact absent from it and check whether the model admits the limitation.

Chat attachments belong to that thread. To make a document available across workspace threads, use workspace embedding instead. Do not assume uploading a file teaches the model permanently or makes it visible in every conversation. Current document-chat guide.

3. Keep privacy and recovery deliberate

Review the language model, embedding model and vector database separately. A local desktop interface can still send content to a selected cloud service. The app also has a telemetry control. Removing a document from one workspace is not full deletion: deleting it from My Documents removes its parsed content and cached embeddings across workspaces. Data-handling details.

Before updating, open the app's storage folder through the version-number menu, quit the app, and back up that storage plus your original documents. Use the same installation method to update; official Windows and Mac installers preserve application storage, while Linux replaces the AppImage. Find your data ยท Update instructions.

If answers ignore the file: check the correct thread and attachment, try a shorter document and ask a specific question before tuning retrieval. If the app cannot reach a local provider, make that provider answer a simple prompt independently first.

Official references

Choose another guide