Meet Hudini, Hudu's AI assistant. Hudini v2 rebuilds the assistant into a persistent, agentic chat you can keep open while you work — it can read and write real Hudu data with your own permissions enforced, find articles and processes by meaning with semantic search, and pull in the records, pages, and files you tag as context. Configure it with your preferred AI provider to get started.
Visit Hudu AI Principles for additional information on how Hudini is designed with safety and ethics in mind.
Supported providers
- Open AI
- Azure OpenAI
- Grok
- Claude
What you will need
- Admin or Super Admin to toggle Hudini on
- For Open AI, you will need: API Key, Model
- For Azure OpenAI, you will need: API Key, Endpoint, Deployment Name, API Version
- For Grok, you will need: API Key, API Endpoint (https://api.x.ai/v1), Model
- To use semantic search, you'll need Open AI or Azure OpenAI configured as your embeddings provider
Hudu does not manage or include AI provider usage. You will need to connect your own account with the provider you choose and have API credits or billing set up separately.
Getting started
To start using Hudini:
- An Admin will need to turn it on in Admin > Hudini.
- Once it's turned on, an AI icon appears in the top nav for everyone with access.
- Choose your preferred provider and enter its credentials, then select Save.
Configuring your provider
Navigate to Admin > Hudini, enable Hudini, and select Open AI from the provider list. Enter your API Key and Model, then select Save.
What Hudu needs from Azure OpenAI: Endpoint, API Key, Deployment name.
In Azure:
- Navigate to Azure Open AI, click +Create (or +Create Azure OpenAI), then click Next.
- Configure network security for your Azure AI services resource and click Next.
- Add tags (optional) and click Next.
- Review and click Create.
- Click Go to Resource, then select Explore Azure AI Foundry Portal.
- Select Create new deployment, choose From base models, select a model, and click Deploy.
Navigate to Shared Resources > Deployments on the left panel to copy your API Key and find your endpoint.
In Hudu:
- Navigate to Admin > Hudini, enable Hudini, and select Azure OpenAI from the list.
- Enter your API Key and Endpoint from your deployment. The endpoint needs to be edited to look like: https://example.openai.azure.com/openai/deployments/deployment-name (the deployment name should match the model selected for the deployment).
- Enter the API Version, found at the end of the Target URI in the Endpoint box.
Navigate to Admin > Hudini, enable Hudini, and select Grok from the provider list. Enter your API Key, API Endpoint (https://api.x.ai/v1), and Model, then select Save.
Navigate to Admin > Hudini, enable Hudini, and select Claude from the provider list. Enter your API Key and Model, then select Save.
Chatting with Hudini
Click AI in the top nav, or press Shift+H, to open the Hudini dock. If a supported record or article is open when you do, Hudini automatically picks it up as context.
- Drag the minimized Hudini shortcut to any edge of your browser window and drop it there.
- Minimize the panel instead of closing it so you can keep following along with a set of instructions while you work, or close it completely and reopen it any time by clicking AI in the top nav.
- Start a new chat with the + control inside the panel. Chats are saved automatically, listed in a switcher so you can pick up where you left off, and auto-titled after your first message.
- Edit a message you already sent, the same way you would in most chat apps.
- While Hudini is working, you'll see a live status line and a Stop button if you want to cancel the run.
- Inactive chats are cleaned up automatically after a month with no activity.
When Hudini needs more detail before it can help — for example, before creating a new article — it walks you through a short set of clarifying questions ("1 of 5") using an interactive card instead of freeform back-and-forth. Your progress is saved if you navigate to another page while it's open.
Adding context to a chat
- Type @ in the composer to tag a company, asset, password, or other supported record and pull it in as context.
- Attach files directly to a message — images (for vision-capable models), PDFs, Word documents, and other text-readable files. Files aren't stored permanently; only the extracted content is kept with the chat.
- From a record's page, use "Add this page to your AI chat" to attach it as a removable chip.
Using AI Search
If you type a question into quick search or advanced search that isn't a match for standard search (for example, "what is Jordan's phone number"), select the AI Search button next to the search bar. It opens a new Hudini chat and sends what you typed straight in as a question — Hudini can then look up the answer across your records.
Skills (formerly Prompt Rules)
Skills are reusable, pre-written prompts you can apply to a message instead of typing the same instructions every time. Prompt Rules from earlier versions of Hudini carry over automatically as Skills.
- To add a new skill, select New Skill, name it, and enter the prompt body.
- Only Admins can decide whether a skill is shared with everyone or kept personal.
- Apply a skill to a message by typing / in the composer and selecting it from the list.
Semantic search (RAG)
Semantic search lets Hudini find knowledge base articles and processes by meaning, not just by matching keywords in the title. Once enabled, Hudini can locate and cite relevant content even if your question doesn't share any wording with the article itself.
- Semantic search requires Open AI or Azure OpenAI configured as your embeddings provider — it isn't available with Grok or Claude as the embeddings provider.
- An Admin enables it and can trigger a search index rebuild from Admin > Hudini. Existing content isn't embedded automatically, so a rebuild is needed the first time it's turned on, or after changing embeddings settings.
- If embeddings aren't configured or become unavailable, Hudini falls back to name-based article search automatically.
- Self-hosted instances need Postgres 18+ with the pgvector extension (.8+) before semantic search can be turned on.
Visit Self-Hosted Upgrading PostgreSQL for steps on upgrading Postgres and pgvector on a self-hosted instance.
Additional Abilities
Admin AI settings
The Admin > Hudini page controls your chat provider, embeddings provider (used for semantic search), the RAG (semantic search) toggle, and the MCP toggle, along with a manual "rebuild search index" action.
Restricting access by group
Hudini access is on for the whole instance by default, but Admins can turn it off for specific groups from the group's Access tab. Members of a restricted group lose all Hudini entry points — the AI icon, the dock, Shift+H, and AI Search. Admins and Super Admins keep access to Hudini even if their groups have it turned off.
Citations and grounding
Every record link Hudini includes in a response is checked against what it actually found while answering. Links that don't check out are corrected or removed, so the sources shown alongside a message are real.
Keyboard shortcuts
- Shift+H — toggle the Hudini dock open or closed
FAQ
Prompt Rules are now called Skills. Existing Prompt Rules carried over automatically and work the same way — apply one to a message by typing / in the composer.
Semantic search needs Open AI or Azure OpenAI set as your embeddings provider. Your chat provider can still be Grok or Claude — the embeddings provider is configured separately.
Yes. An Admin can turn Hudini off for a group from that group's Access tab. Admins and Super Admins always keep access, regardless of their group settings.
No. Attached files aren't persisted — only the text or data extracted from them is kept with the chat.
Hudu doesn't manage or include AI provider usage. You'll need your own account and billing set up with whichever provider you choose to use with Hudini.
Troubleshooting
If RAG isn't available on a self-hosted instance (for example, before upgrading Postgres/pgvector), it turns itself off cleanly and Hudini continues to work with name-based search — no errors, just less precise results.
Let me know if you'd like to adjust any sections, add screenshots, or refine the copy.