Analytics cookies only (Google Analytics, PostHog). No ad networks, no cross-site tracking. See our privacy policy.
Ma
Self-host MaxKB
updated Nov 2026deployed by usprices checked · Jul 2026Install verified on Ubuntu 26.04 · Nov 2026 · how we test
We earn commissions when you shop through the links below. Full disclosure →
A platform for building question-answering agents over your own documents. You upload files or crawl web pages, MaxKB splits the text and stores vectors for retrieval, and agents answer from that knowledge base using the model you connect. It also has a workflow editor, a function library, MCP tool use and multimodal input and output, and a chat can be embedded into another system without code. It supports private models (DeepSeek, Llama, Qwen) and public ones (OpenAI, Claude, Gemini, MiniMax). The Docker image runs the web app, a task worker and PostgreSQL with pgvector in one container.
Pick MaxKB when the job is a question-answering assistant over your own documents and you want it running from a single `docker run`, with nothing else to wire up. You upload files or crawl web pages, it splits and vectorizes the text, and an agent answers from that knowledge base using the model you connect; a workflow editor, a function library and MCP tool use cover the cases where plain retrieval is not enough, and a chat can be embedded in another site without code. The image bundles the web app, a worker, PostgreSQL with pgvector and Redis, so there is no compose file to maintain. It is GPL-3.0 (the repository's licence file is the unmodified GPLv3 text), Python and Django with a Vue front end. Our snippet adapts the README quick start, pins `1panel/maxkb:v2.10.8-lts` (pushed to Docker Hub 2026-10-22, amd64 and arm64) instead of the unpinned tag the README uses, and keeps the data in a named volume. The official installation docs (docs.fit2cloud.com/maxkb/installation/online_installtion) state a minimum of 4 cores and 8 GB of RAM, so our floor is 8192 MB; a local run held about 2.5 GB with nothing configured. Among the catalog's agent platforms, Dify is the larger builder with a separate service stack, and Flowise and Langflow are visual node builders; MaxKB is the one to take when the knowledge base is the product rather than the number of workflow nodes.
What you need
Any VPS with at least 8192 MB of RAM
A domain you control — most self-hosted setups need HTTPS in front of them
Paid link — we earn a commission if you shop through it.
Install
Run these commands on your server:
# MaxKB — RAG and agent platform — the README's single-container quick start, pinned to v2.10.8-lts (the image bundles PostgreSQL with pgvector and Redis, so there is no compose file)docker run -d --name maxkb --restart=always -p 8080:8080 -v maxkb_data:/opt/maxkb 1panel/maxkb:v2.10.8-ltstimeout 300 bash -c 'until curl -fsS -o /dev/null http://localhost:8080/admin/; do sleep 3; done'# → http://SERVER_IP:8080 — the MaxKB login page; the README's default account is admin / MaxKB@123.. so change it straight away
MaxKB's README and image flag these before you rely on it:
non-negotiableThe README publishes a default administrator account, admin with the password MaxKB@123... Anyone who can reach port 8080 can sign in with it until you change the password, so do that on first login and keep the port off the public internet until you have.
non-negotiableAll state, including the bundled PostgreSQL database and uploaded documents, lives in the one /opt/maxkb volume. Back that volume up before changing the image tag, and treat the single container as the whole database server: there is no separate database to scale or snapshot.
non-negotiableNothing answers questions until you add a model. MaxKB does not ship a chat model (the image bundles only a local embedding model), so connect a hosted provider with an API key or a local model server in the settings before the first agent will respond.
non-negotiableThe tag we pin is an -lts line, and the Docker Hub list also carries a v1.10.15-lts line from August 2026 alongside the v2.x one. The two are separate major lines: v1 to v2 needs the separate migration tool, and the v2 image refuses a v1 data directory. Do not use the floating latest or v2 tags in production.
non-negotiableThe README links an offline installation guide because pulling the image fails for some users in China. The image is about 1.2 GB compressed on amd64, so plan for the first pull and for the disk it takes.
MaxKB used 1732 MB of RAM at idle on GCP e2-standard-2 · Ubuntu 26.04 LTS · Docker 29.9.3 (first boot, measured Nov 2026). We list a minimum of 8192 MB.
Is the MaxKB install snippet verified?
The MaxKB install snippet is marked verified on Ubuntu 26.04 (Nov 2026).