Ranked shortlist
5 best open-source AnythingLLM alternatives
Most teams start shopping for an open-source AnythingLLM alternative when private document chat stops being the whole job. AnythingLLM is a local-first app, MIT-licensed, for chatting with your documents and running agents on your own hardware, and it does that job well.
The five at a glance
| Tool | Open source | Best at | Where it stops |
|---|---|---|---|
| Kortix Recommended | Yes | Agent work across tools, governed, in a repo you own | A platform to run, not a chat window |
| Open WebUI | Custom licence | Self-hosted chat over local and cloud models | Chat-first work model |
| LM Studio | No | Local models on one desktop, fully offline | One machine, no team layer |
| LibreChat | Yes | One team interface over many providers | Chat-first, experimental agent workspaces |
| AnythingLLM | Yes | Private document chat in one app | Single-app scope |
The limits show up with scope. It is one app around one workspace of documents, while company-level agent work needs reach across tools, state that survives between runs, isolation for each task, and rules about who may act. The ranking below uses four criteria: what you own, what the agents can actually do, what governance exists, and how you host it. On that standard, Kortix is the recommendation. Open WebUI, LM Studio, LibreChat and AnythingLLM itself cover the smaller jobs, and each gets a fair look.
1. Kortix: the open-source AI Operating System
Kortix is the open-source AI Operating System: your agents, their skills, your company memory, and every connector in one git repo you own, with the agents working on real cloud computers. The repo is the product. Agents are markdown files, skills are markdown plus scripts, memory is files that accumulate, and kortix.yaml declares the connectors, triggers and permissions. Every part of it is text you can read, diff and version, and you can grep the whole company (Kortix on GitHub).
Work runs as sessions. Each session gets its own isolated sandbox on its own branch, so an agent can install, run and break things while only what it commits survives. Agents reach 3,000+ apps in a click, plus MCP, OpenAPI, Postman, GraphQL and raw HTTP, and you set every tool to Allow, Ask or Block, down to the arguments of each call. Nothing merges itself: work reaches main only through a change request a person approves, and merge is default-deny for agents.
Model and hosting choice stay yours: any model provider with your own keys, your own models behind your own URL, or the ChatGPT plan you already pay for. Run it on Kortix Cloud, in your VPC, or on your own on-prem network, and self-host for free.
Where it stops: Kortix is a platform to run and govern, so it asks more of you than a chat app does. Its own README points readers who only want a chat window over their documents to Onyx, Open WebUI or LibreChat instead. For the direct comparison, see the Kortix vs AnythingLLM breakdown, and for hosting specifics see the guide to self-hosting an AnythingLLM alternative.
2. Open WebUI: a self-hosted chat interface for local and cloud models
Open WebUI is a self-hosted AI interface that supports Ollama alongside any OpenAI-compatible API, built to run entirely offline. Its strength is the chat surface itself: granular role-based access with user groups, local retrieval-augmented generation backed by a choice of 9 vector databases, and a plugin system that adds filters, tools, skills and MCP tool servers. Agentic execution exists through Open Terminal, which gives an agent a terminal and a filesystem for multi-step tasks, and the enterprise edition isolates those environments per user.
One licensing detail matters for companies. Open WebUI ships a custom licence that permits self-hosting but prohibits altering or removing its branding unless a deployment stays at 50 or fewer end users per rolling 30 days, or you obtain written permission (checked October 2026).
Where it stops: the center of gravity is chat. Open WebUI fits teams that live in threads, document question-answering and model comparison. Governed agent work across your business tools, with review before anything lands, is not its core model.
3. LM Studio: local models on your desktop
LM Studio is a desktop app for macOS, Windows and Linux that downloads and runs local models such as Llama, Qwen, gpt-oss and DeepSeek R1 on llama.cpp (GGUF) runtimes, or MLX on Apple Silicon. It is the most direct path to private, offline use: chat with documents entirely offline, connect MCP servers to local models, and serve local models on OpenAI-compatible endpoints to other apps on your network. A headless variant named llmster runs the same engine on servers and in CI.
Two things to know before you pick it. The app itself is closed software, distributed under the site's terms of use rather than an open-source licence, and its homepage now leads with Bionic, an agent for work and code (checked October 2026). Where it stops: LM Studio serves one machine and one person's work. A team that wants shared agent state, connectors into business tools and permission rules has to look past it.
4. LibreChat: a multi-provider chat platform for teams
LibreChat is a self-hosted, MIT-licensed chat platform that puts Anthropic, OpenAI, AWS Bedrock, Google Vertex AI, Ollama and any OpenAI-compatible endpoint behind one ChatGPT-style interface. Teams pick it for the admin layer: an admin panel, OAuth2 and LDAP sign-in, moderation and token spend controls, and multi-user access out of the box. Its agents support MCP tools, reusable SKILL.md skill bundles, subagents, and a sandboxed code interpreter that runs Python, Node.js, Go, C/C++, Java, PHP, Rust and Fortran.
Where it stops: LibreChat is chat-first. Attached code workspaces, where an agent inspects and edits files in a repo, carry a highly experimental label in its own changelog, and there is no layer that turns agent output into reviewed changes to a company repo.
5. AnythingLLM: still the right tool for private document chat
AnythingLLM earns its place here because the switch is usually about scope, not quality. It is open source under the MIT licence, it runs locally by default, and it bundles document chat, built-in agents, scheduled tasks on a cron schedule, MCP compatibility and dynamic model routing into one application. Desktop builds cover Mac, Windows and Linux, and the Docker deployment adds multi-user accounts with permissions plus an embeddable website chat widget; as of October 2026, both features are Docker-only.
Where it stops: the single-app boundary. AnythingLLM's model is one app, one workspace, one deployment. When the job becomes agents working across many tools with shared company memory, isolation between runs and rules about who may act, that is where the alternatives above take over.
How to choose
Match the tool to the job you actually have. For private document chat by yourself on one machine, LM Studio is the simplest answer, and AnythingLLM remains a good one. For a team that mostly asks questions over documents and local models, Open WebUI provides that surface with roles and groups. For one interface over several paid model providers with an admin layer, pick LibreChat. Pick Kortix when the job is agent work: agents that take actions in your tools, run on schedules, keep company memory, and land their changes for a person to review. That is a different job from document chat, and the tools are built differently because of it. The open-source AnythingLLM alternative overview compresses the same decision into one screen.
If that last job describes you, try Kortix and judge the change-request loop yourself. It is open source, so you can read every part of it first.
Frequently asked questions
Which open-source AnythingLLM alternative is completely free?
The four open-source tools are all free to self-host. Kortix is open source and free to self-host, with a managed cloud option if you would rather not run it yourself; Open WebUI, LibreChat and AnythingLLM install free under their licences. LM Studio is the exception: the desktop app is closed software under its terms of use, though the models you run through it stay on your hardware.
Do these alternatives work with Ollama and local models?
Open WebUI is built around Ollama plus OpenAI-compatible APIs. LM Studio runs GGUF and MLX models on your desktop and can serve them on OpenAI-compatible endpoints. LibreChat supports Ollama and custom OpenAI-compatible endpoints. Kortix takes any model provider with your own keys, and it also runs your own models behind your own URL, so a self-hosted model works there too.
Is AnythingLLM itself open source?
Yes. AnythingLLM is licensed under the MIT licence, and its Android mobile app is open source as well. Licensing is rarely what drives a switch; the scope limits covered above usually are.