BSuit AI and AI Studio: set up, train, chat and extend
Connect an AI provider, feed it your data, control what the AI user may do, use chat with confirmed actions, and understand how AI Studio’s Developer tools relate to it.
On this page
Process at a glance
- Choose the providerAdministratorA local model or an external API provider is connected and tested.
- Collect and embed documentsAdministratorERP data is collected as documents and embedded for semantic search.
- Train the local model (optional)AdministratorModules are chosen, data initialised and a training run completed.
- Limit the AI userAdministratorCapability, risk limit and an action allow-list are set.
- Ask and confirmAny userAnswers about business data; any proposed action waits for Confirm.
- Extend BSuit with Developer toolsAdministratorCustom entities, forms, rules and change requests (AI Studio).
#Purpose and scope
BSuit AI is the assistant: provider settings, documents, training, an AI user identity, chat and insights. AI Studio is the installable module behind the Developer menu: custom entities, forms, business rules and change requests. In the app, BSuit AI appears as the Intelligence lane under the single AI Studio heading, but the two do different jobs. If AI Studio is not installed, BSuit AI stays reachable on its own.
#Before you start
- The BSuit AI module is installed for your company. If no configuration exists, the settings page tells you to install it to initialise settings.
- For an external provider you need its API base URL, API key and model name. The AI settings page has presets for common providers and a Test Connection button.
- Administrator access for settings, documents, training and the Developer menu.
#1. Connect a provider
- Open AI settings. At the top a status chip shows the active connection (for example Connected, Connection failed or API key required).
- On the AI Provider tab choose the mode: Local Model (the built-in BSuit-AI model, which requires training) or API Provider (an external API such as OpenAI, Anthropic or Ollama).
- For an API Provider, pick a preset or enter API base URL, API key, model, temperature, max tokens and system prompt, then choose Test Connection.
- On the RAG & Embeddings tab set whether RAG is enabled, the number of results (top K), and the embedding provider and model. Text-to-SQL is an advanced option there.
- On the Local Model tab set model size, sequence length, training epochs, batch size, learning rate, auto-collect interval and the generation limits.
#2. Collect and embed documents
- Open AI documents. Choose Collect Now to gather data from ERP modules; the toast reports how many new documents were collected.
- Filter by module or status (Pending, Embedded, Failed) and search the content. Each row shows the content preview, module, entity, token count and status.
- On the settings page, RAG & Embeddings tab, choose Embed Documents to embed all collected documents into the vector database so chat can search them semantically.
#3. Train the local model (optional)
- Open Training. On the Training tab select modules, review the record counts, and set epochs, batch size, learning rate and records per entity.
- Choose Initialize to collect the data. The run is shown as initialised; you can then Start Training or Discard it.
- While a run is in progress you see the epoch, loss, step and per-module progress. You can Pause (progress is saved), Resume, or Cancel (collected data is kept).
- The Plain Text tab imports custom text (policies, FAQs, product descriptions). The Documents tab uploads PDF, Word, Excel, TXT or CSV files, up to 50 MB each.
- Training history lists each run with its version, losses and start time; select runs to remove them.
#4. Set what the AI user may do
AI user controls the non-human identity the assistant uses to act. Both the signed-in human and this identity must permit an action.
| Setting | Meaning |
|---|---|
| Enabled | Switches the AI identity on or off. When off, no AI action can be prepared. |
| Display name | How the identity is shown. |
| Capability level | 1 Observe, 2 Draft, 3 Propose, 4 Execute approved, 5 Autonomous low-risk. Level 1 is observe-only and cannot prepare actions. |
| Maximum action risk | Levels 1 to 6. Actions above this level are refused. |
| Workflow required from | From this risk level up, the identity must be at capability level 4 or higher to run the action. |
| Action allow-list | No action is available until it is ticked. Each action shows its module and permission key. |
#5. Chat and confirm actions
- Open Chat, choose New Chat and ask about your business data (invoices, products, employees, reports). Past conversations are listed on the left and can be deleted.
- If your message asks for an action, the assistant shows a proposal card. Nothing is written until you choose Confirm; Cancel discards it.
- On Confirm, BSuit executes the action server-side with your own permissions.
#6. AI Insights
The BSuit AI home shows the module’s pages and an AI Insights grid. Run any module’s insights to render a chart from your real data, and export all of them as one PDF.
#7. AI Studio: the Developer tools
AI Studio provides the Developer menu, visible to administrators only. If the menu is missing, the module is probably not installed for your company.
| Page | Use it to |
|---|---|
| Change requests | Package structural changes, then release, apply or roll them back. |
| Applications | Define low-code applications that group entities and screens. |
| Entities | Create custom entities and inspect built-in ones. |
| Form builder | Lay out and configure forms for custom and standard entities. |
| Business rules | Attach organisation-specific logic and validation to records. |
| Form actions | Configure actions on forms. |
| Notifications | Notify users when a record changes. |
| Audit config | Set per-entity audit policy and excluded fields. |
#Who can do what
Anyone signed in can use chat and their own conversations. Settings, documents, training, the AI user, knowledge sync and the Developer menu are administrator tools. The AI actions feature can be switched off per company; when it is, the actions endpoints respond as not found.
About this guide
- Status
- Partially reviewed
- Application
- BSuit ERP
- Audience
- Administrators, developers and users of the AI assistant
- Owner
- BSuit documentation team
- Reviewer
- Not yet reviewed
- Last verified
- Never
- Checked against
- Generated from source revision a36c7a086 on 2026-10-02
- Seen on a running system
- Not yet
- Help version
- 2026.0.0-preview
- Guide ID
erp:article/guide/bsuit-ai-studio