Multi-LLM for TYPO3: Why You Shouldn't Commit to a Single AI Provider

When an AI feature in TYPO3 is tied to one provider, your website changes whenever that provider retires a model or rewrites its terms. Multi-LLM for TYPO3 removes that dependency: you connect several language models from different providers to one TYPO3 installation and decide which model handles which job. A large language model, or LLM, is the technology behind services such as ChatGPT and Claude.

With AI Foundation for TYPO3, T3Planet's free and open source base, that is a backend setting rather than a rebuild. This guide covers why one provider is a risk, how Multi-LLM works under the hood, and how to add a second provider in four steps.

Multi-LLM for TYPO3 means your installation runs several AI models from different providers side by side, and you decide in one place which model does which job. Your extensions never call OpenAI, Anthropic or Mistral directly. They call a shared layer in between, so changing the model leaves their code untouched. For you, choosing a provider stops being a permanent decision.

This is vendor lock-in for AI, and in TYPO3 it affects every feature that depends on a single model. None of the reasons below is a criticism of any provider. They follow from how fast the AI market moves, and for you they mean one thing: hard-wire a model today and you are already scheduling the next change.

1. Models come with an end date

Anthropic announced the retirement of Claude 3.5 Sonnet in August 2025 and switched it off on 28 October 2025. Its model deprecation overview commits to at least 60 days' notice for publicly released models. OpenAI's deprecations list gives generally available models at least six months, and the August 2025 GPT-5 snapshot shuts down on 11 December 2026.

If the model is baked into your code, every retirement becomes a development ticket. In AI Foundation you select the successor in the backend, and every extension using that provider works with it straight away, unless a feature pins a model of its own.

2. The provider sets the prices and the terms

Token prices, usage limits and contract terms are the provider's call, not yours. With only one provider connected, a change leaves you little choice but to accept it. With a second provider already tested, you have a real alternative, and your own dashboard gives you the numbers to weigh it.

3. No model is best at everything

A translation asks different things of a model than alt text for an image or a semantic search does. Images need a model that can see. Search needs an embedding model, which turns text into numbers that can be compared. Specialist services such as DeepL for translation or AltText.ai for image descriptions connect to AI Foundation alongside the large language models.

4. Not every piece of content should take the same route

A public blog post is not the same as an internal draft that mentions customers. A model hosted locally or in the EU keeps data in that jurisdiction; a provider outside the EU does not. With Ollama you run open source models on your own server, so the content never leaves your infrastructure. Together with your data protection team, you decide which content takes which route.

You manage every AI provider once, centrally, and every AI extension on your TYPO3 installation uses them. AI Foundation sits between your extensions and the providers. Your code never talks to a provider directly, only to one AI Foundation interface. Our extensions work the same way, so they fit into your existing TYPO3 setup instead of bringing their own provider keys. Underneath is Symfony AI Platform, a Symfony component whose documentation describes switching between models and providers without changing application code.

The providers you can connect

What you set for each provider

Each provider is its own entry under AI Providers. You can add the same provider more than once, for example with one key for production and another for staging.

• Models: a chat model, picked from a list fetched from the provider, plus an embedding model of its own if you want one
• Capabilities: chat, completion, embeddings and vision, each switched on separately
• Behaviour: temperature and a system prompt of your own
• Cost: input and output price per million tokens, currency and cost centre, so the dashboard can work out what each provider costs
• Logging: standard, reduced (no prompt content) or off
• Security: API key stored encrypted, one-click connection test, and settings importable from your other TYPO3 sites

Which provider answers

AI Foundation picks the provider in this order:

1. the provider an editor selects in the AI dialog
2. the default for that feature, set under AI Features
3. the global default provider under AI Providers

Per-feature settings appear once an AI extension registers them. AI Assistant brings its own defaults for SEO, pages, content and translation. AI Chatbot and AI Search let you choose their own chat and embedding provider. That way translations can run on Mistral, for example, while metadata comes from OpenAI. These extensions are premium products with their own licence; AI Foundation itself is free.

When the model changes, everything you have set up for your website stays in place, because those rules live in AI Foundation and not with the provider.

• Brand voice: your brand profile under AI Context, with tone, audience, content rules and banned words, goes into every AI request, whichever model answers
• Prompts: the prompt templates of every AI extension sit centrally under AI Prompts and apply to every provider
• Permissions: under AI Permissions you decide, per backend user group, which providers are allowed, whether the model may be overridden and how many requests are possible per day. The values are written into native TYPO3 group permissions
• Audit trail: depending on the logging level, each request is logged with time, user, extension, feature, provider, model and tokens

The dashboard shows cost and error rate per provider, plus the most used models. That is how you manage AI models in TYPO3 in one place rather than extension by extension, and how you compare two providers on your own traffic before you and your team change the default.

AI Foundation is free and open source, with no licence key. One Composer command installs it, and Quick Setup walks you through the first provider.

Install AI Foundation for free

Until you make the new provider the default, your existing features carry on unchanged. So you and your team can test at your own pace while editors keep working as usual.

Step 1: Add the provider

• In AI Foundation, open AI Providers and click New Provider.
• Give it an identifier and a display name, for example mistral-prod and "Mistral production".
• Choose the adapter and enter the API key. For a local Ollama model, leave the field empty.

The key is stored encrypted in the database.

Step 2: Set models and cost

• Fetch the model list and choose the chat model.
• Set an embedding model, or leave the field empty to use the chat model for that too.
• Enter the price per million tokens and a cost centre.

The dashboard can now work out what this provider costs.

Step 3: Test and assign it

• Check the entry with the connection test.
• Use the star to make it the global default, or assign it to a single feature under AI Features.
• Editors can also pick it in the AI dialog.

Your extensions pick up the new assignment with no further setup.

Step 4: Control access and watch the numbers

• Under AI Permissions, decide which user groups may use the provider.
• Set a daily request limit if you need one.
• After a few days, compare cost and error rate on the dashboard.

Then decide which provider takes which job for good.

Note: If you change the embedding model behind a search, rebuild the search index afterwards. Vectors from different embedding models cannot be compared.

Not every team wants contracts with several AI providers and a set of API keys to look after. That is what AI Credits is for: an optional prepaid balance from T3Planet that runs the AI features of the AI Universe without a key of your own. Your own key stays the default, and we support you on either route.

• Start with a name and an email address: no API key, no licence key. To try it, you get 50 free credits once per account.
• One balance across extensions: it covers every AI Universe extension and every installation, including development and staging. Only voice features keep running on your own key.
• Usage, not a licence: credits pay for usage, never for a licence. Premium extensions such as AI Assistant still need their own.
• Cost visible in advance: before each action the backend shows an estimate in credits, and every action is logged with its credits.
• No overage bill: there is no auto-recharge. The most you can spend is what you have prepaid. Plans start at €15 a month.

The trade-off against Multi-LLM on your own key: T3Planet picks the models, and your requests reach the model providers through T3Planet. Need a data processing agreement that covers AI Credits? Ask and we will send it. Switch the mode under AI Providers, in either direction, whenever you like. Our post on AI costs in TYPO3 compares both routes.

Try AI Credits for free

Whether you use AI Foundation or are evaluating something else, these four questions show quickly how dependent on one provider you would become. Here are our answers for AI Foundation.

One line is all it takes, and you can connect your first and second AI provider today:

The source code is on GitHub, and every feature is on the AI Foundation for TYPO3 product page. AI Foundation is built by NITSAN: more than 15 years in TYPO3, a Gold Member of the TYPO3 Association, winner of the TYPO3 Award 2018 and publisher of more than 100 open source extensions. If you would like to work through your provider strategy together, book a demo with Jürgen, our AI Consultant for TYPO3.