Every AI your people use
Your people use a dozen different AI tools. Axonyx applies one set of rules to all of them, whether the model runs in the cloud or on your own hardware.
The full list
| Provider | Used for | Models covered | What you get |
|---|---|---|---|
| OpenAI | Text (Chat / API) | GPT-4o, GPT-4.1, GPT-4 Turbo, GPT-3.5 Turbo | Every question and answer is checked. Risky ones are blocked or sent for approval. All of it is recorded. |
| OpenAI | Text-to-Speech (TTS) | GPT-4o TTS, GPT-4o Mini TTS | Text is checked for customer or staff data before it becomes speech. |
| OpenAI | Image Generation | GPT-4o Image, DALL·E | Image requests are checked for misuse and kept on record. |
| Anthropic | Text (Chat) | Claude 3 Opus, Claude 3.5 Sonnet, Claude 3 Haiku | Questions and answers are checked both ways, with a record of each one. |
| Google (Gemini) | Text / Multimodal | Gemini 1.5 Pro, Gemini 1.5 Flash | Text, images and files are all covered by the same rules. |
| Mistral AI | Text (Chat / Completion) | Mistral Large, Mixtral 8×7B, Mixtral 8×22B | Same checks and same records as the bigger providers. |
| Meta (Llama) | Text (Self-hosted / Cloud) | Llama 2, Llama 3, Llama 3.1 | Works the same whether Llama runs in the cloud or in your data centre. |
| DeepSeek | Text / Code | DeepSeek-Chat, DeepSeek-Coder | Catches source code and secrets before they leave your business. |
| Microsoft (Azure OpenAI) | Text | GPT-4 / GPT-4o, GPT-35 Turbo | Runs inside your private network, with records built for auditors. |
| Amazon (Bedrock) | Text / Multimodel | Claude (Bedrock), Amazon Titan, Mistral | One set of rules across every model you run in Bedrock. |
| Cohere | Text / RAG | Command, Command R / R+ | Covers answers drawn from your own documents, not just the model. |
| IBM (watsonx) | Text | Granite series | Built for regulated industries that keep AI on their own hardware. |
| Open / Self-Hosted | Text / Multimodal | Any OpenAI-compatible or open-source model | Any model you build or host yourself gets the same treatment. |
| Browser & Agents | Web / Agentic AI | ChatGPT, Claude.ai, Gemini UI, copilots, agents | Covers staff using AI in a browser, and AI agents acting on their own. |
Questions we get about model and provider coverage
Which AI models does Axonyx support?
The major commercial providers, including OpenAI, Anthropic, Google, Meta, Mistral, Cohere and AWS Bedrock, alongside open models you host yourself.
The point of the design is that the governance does not live with the model. Adding a provider does not mean rebuilding your policies.
Can we use our own self-hosted models?
Yes. Models you run in your own infrastructure sit behind the same gateway and carry the same rules and the same evidence as a commercial API.
For teams whose reason for self-hosting is data residency, this matters: you keep the model in your estate and still get the record.
What happens when a new model is released?
It is added centrally, and you choose whether to allow it. Nothing appears in front of your staff because a provider shipped it.
This is the practical reason to put a gateway in the path. Model releases are frequent, and each one is otherwise a small governance decision somebody has to remember to make.
Can we route different work to different models?
Yes, and most teams end up doing so. You can route by cost, by sensitivity, by team, or by which provider is approved for a category of data.
A request touching customer records can be pinned to an approved model while general questions go to a cheaper one.
See it running on your own AI.
See your own AI traffic, and what Axonyx would have stopped.
Get Started