The self-hosted version is, and it is functionally unlimited: MIT-licensed, run through Docker, with no stated functional limits. What bounds it is your own hardware and the model key you bring. The paid tiers are the managed cloud — Basic at $50 a month, Pro at $99, and Enterprise quoted.

Do I still need my own API key on the paid plans?

Yes, on Basic and Pro. The pricing page lists your own LLM key as a requirement on Basic, and Pro carries it too, since Pro is Basic plus those two additions — large-team support and a 72-hour SLA. Enterprise is quoted rather than listed, so the pricing page does not say either way there. The subscription pays for the private instance, the custom subdomain, RAG and agents — not for the model doing the answering.

Why are the answers inconsistent?

Usually the documents. A review of AnythingLLM found that RAG quality depends entirely on source hygiene, and that conflicting documents produce fluctuating answers. Narrowing and de-duplicating the source set does more here than swapping models.

What does self-hosting actually cost me?

Not money. The same review is blunt about the trade: you own updates, patching, backups and access control. If nobody wants that job, the $50-a-month managed instance is what you are really buying.

Is the project active?

65,700 GitHub stars, last pushed on 4 September 2026, under an MIT licence. Release v1.16.1 on 27 August 2026 added Foundry Local for Windows GPU and NPU hardware, plus Qualcomm GenieX for Snapdragon NPUs.

AnythingLLM: pricing, limits, alternatives (2026) • SaikoDev

Good for / not for

Individuals and small organisations who want private document chat without assembling a stack themselves.
Anyone who already pays for a model API: you bring your own key on Free, Basic and Pro, whether self-hosted or managed.
People who want a clean, permissive licence with no branding or use restrictions — MIT, at 65,700 GitHub stars, last pushed on 4 September 2026.
Recent Windows and Snapdragon hardware: v1.16.1 on 27 August 2026 added Foundry Local for Windows GPU and NPU, and Qualcomm GenieX for Snapdragon NPUs.

Organisations wanting turnkey SaaS with white-glove support.
Messy document sets. RAG quality depends entirely on source hygiene, and conflicting documents produce fluctuating answers.
Anyone who does not want to own updates, patching, backups and access control, because that is what self-hosting means here.

What it actually does

AnythingLLM is a workspace for talking to your own material. You point it at documents, it indexes them, and a model answers from those rather than from whatever it absorbed in training. The same workspaces host agents. It is MIT-licensed and self-hosted through Docker, with no stated functional limits — what bounds you is your hardware and the model key you bring. That holds on Basic and Pro too: AnythingLLM supplies the app and the retrieval, not the model, so a key of your own is part of the setup on those paid plans as well. The managed cloud is the alternative to running it yourself. Basic is $50 a month for a private instance with a custom subdomain, RAG and agents. Pro is $99 and adds large-team support and a 72-hour support SLA. Enterprise is quoted. The project is moving: 65,700 GitHub stars, last pushed on 4 September 2026, and v1.16.1 on 27 August 2026 added Foundry Local for Windows GPU and NPU hardware plus Qualcomm GenieX for Snapdragon NPUs. The honest caveat is that answer quality tracks source hygiene — feed it contradictory documents and the answers fluctuate.

What the free tier is actually bounded by

Self-host AnythingLLM with Docker.

No stated functional limits. What bounds you is your own hardware and the model key you supply — plus the fact that updates, patching, backups and access control are now yours.

From the vendor's docs

Pricing and free-plan limits

Free tier: Self-hosting is functionally unlimited: no stated limits, bounded by your own hardware and the model key you bring.

Plan	Price	What you get
Free	Free	Self-hosted through Docker No stated functional limits Bring your own model key Limits: Bounded by your own hardware · You own updates, patching, backups and access control The free tier is the self-hosted product, not a capped version of the cloud one.
Basic	$50/month	A private managed instance with a custom subdomain RAG and agents Limits: Your own LLM key is still required Managed cloud. The subscription buys the hosting, not the model.
Pro	$99/month	Everything in Basic Large-team support A 72-hour support SLA Limits: Your own LLM key is still required, since Pro is Basic plus those two additions Managed cloud.
Enterprise	Custom pricing	Quoted rather than listed No figure is published.

Prices checked on Sep 6, 2026 against the vendor's pricing page. Plans change often — confirm before you pay.

Three honest limitations

RAG quality depends entirely on source hygiene. Conflicting documents produce fluctuating answers, so the work is in curating what you feed it.
Self-hosting means you own updates, patching, backups and access control. The free tier is free of charge, not free of operations.
The model is not included on Free, Basic or Pro. Self-hosted and managed cloud alike require your own LLM key, so the $50 and $99 plans buy hosting rather than inference.

Alternatives

Open WebUI — Choose Open WebUI instead if you need a multi-user ChatGPT-style front end with SSO more than you need a document workspace.
Jan — Choose Jan instead if this is one person on one desktop and you would rather not run a container at all.
LM Studio — Choose LM Studio instead if the job is picking and running models through a graphical app rather than chatting with your own files.

Getting started in 5 minutes

Run the self-hosted build with Docker first. There are no stated functional limits, so the cloud plans are a convenience rather than an unlock.
Have a model key ready. AnythingLLM brings the workspace and the retrieval; the model is yours to supply on Free, Basic and Pro.
Start with a small, clean set of documents. Answer quality tracks source hygiene, and contradictory sources make answers fluctuate.
Decide who owns the box before you invite anyone in — updates, patching, backups and access control come with self-hosting.
Move to the managed cloud only when hosting is the problem: Basic is $50 a month, Pro is $99, and Enterprise is quoted.

AnythingLLM questions

Is AnythingLLM free?: The self-hosted version is, and it is functionally unlimited: MIT-licensed, run through Docker, with no stated functional limits. What bounds it is your own hardware and the model key you bring. The paid tiers are the managed cloud — Basic at $50 a month, Pro at $99, and Enterprise quoted.
Do I still need my own API key on the paid plans?: Yes, on Basic and Pro. The pricing page lists your own LLM key as a requirement on Basic, and Pro carries it too, since Pro is Basic plus those two additions — large-team support and a 72-hour SLA. Enterprise is quoted rather than listed, so the pricing page does not say either way there. The subscription pays for the private instance, the custom subdomain, RAG and agents — not for the model doing the answering.
Why are the answers inconsistent?: Usually the documents. A review of AnythingLLM found that RAG quality depends entirely on source hygiene, and that conflicting documents produce fluctuating answers. Narrowing and de-duplicating the source set does more here than swapping models.
What does self-hosting actually cost me?: Not money. The same review is blunt about the trade: you own updates, patching, backups and access control. If nobody wants that job, the $50-a-month managed instance is what you are really buying.
Is the project active?: 65,700 GitHub stars, last pushed on 4 September 2026, under an MIT licence. Release v1.16.1 on 27 August 2026 added Foundry Local for Windows GPU and NPU hardware, plus Qualcomm GenieX for Snapdragon NPUs.

Sources

Research-compiled, not hands-on tested. Compiled from official docs, the vendor's pricing page and community reviews as of Sep 6, 2026. We haven't tested it hands-on yet.

AnythingLLM