AnythingLLM
hosomaki.anythingllm
A Docker Compose file for AnythingLLM alone, generated from the catalog and checked by the safety inspector. No account needed.
Retrieval-augmented chat workspace bundling document ingestion, embedding, vector storage and multi-provider model connectivity into one container. Answers cite the source passages they were drawn from.
AnythingLLM organises documents into workspaces: uploaded files are chunked, embedded and written to a vector index, then retrieved as context when a question is asked in that workspace. It brokers connections to local runtimes and hosted APIs, so the chat model can be swapped without reprocessing the corpus, and each workspace keeps its documents and history separate from the others. Answers are returned with citations back to the retrieved chunks, which makes a wrong answer traceable to the passage that produced it. A first deployment means selecting an embedding provider and a chat model, creating one workspace, and uploading a handful of files to query.
You know it worked when
- The interface loads and reports a connected embedding provider and chat model.
- A document uploaded to a workspace finishes processing and appears in its document list.
- A question about that document returns an answer carrying at least one source citation.
- Switching the workspace to a second chat model leaves the existing document index intact.
Known sharp edges
- Document ingestion runs synchronously and a large upload blocks the workspace until embedding finishes.
- The embedding model chosen at first ingest fixes the index; changing it later requires re-embedding every document in the workspace.
- Scanned PDFs and image-only files produce empty chunks unless text extraction happens before upload.
- The default deployment runs as a single administrator account with multi-user mode disabled, so anyone reaching the port has full access to every workspace.