01AI and integrations · module
AI assistant in Draivix
An in-app assistant that finds data, drafts changes and saves them once you confirm
- AI conversation
- AI message
- AI assistant preference
- 11
- capabilities
- 3
- record types
- 2
- automations
- 12
- integrations and AI tools
02What the module does
Less routine. More value.
The AI assistant is a chat built right into Draivix that you talk to in plain language. It looks up records, summarizes sales, inventory or production status, and drafts new or updated records for you to confirm before anything is saved. It always runs under the signed-in user's permissions, so it never sees or changes anything you don't have access to.
- Answers without digging through listsAsk a question in plain language and the assistant searches, counts and reads records across any entity (record type) you have access to—no manual list filtering required.
- Nothing changes without your approvalBy default, the assistant never creates, edits or deletes a record directly. Instead, it prepares a draft showing the before-and-after values, and only saves the change once you confirm.
- Straight to the right screenThe assistant can open a page, apply a filter, launch the mass editor or give you a direct link to a record. Instead of clicking through menus, you pick up work right where you need to be.
- Sales pipeline at a single promptThe assistant summarizes open opportunities, leads missing a follow-up and recent activities on an account. Sales reps immediately see what is in progress and where the next step is missing.
- From meeting notes straight to tasksThe assistant proposes follow-up tasks directly from your meeting notes. You simply review and confirm them, ensuring agreed action items don't get buried in a transcript.
- Inventory status in one questionThe assistant summarizes product stock, low-stock items, today's movements and open stock transfers. You get a clear picture of your warehouse in a single answer without switching between views.
- See what is waiting on the shop floorThe assistant summarizes pending operations and open manufacturing and work orders, giving production managers an instant overview of what is up next and what is still unfinished.
- Roll back a bad data importThe assistant shows the status of data imports and can roll an import back. If an import loads something unexpected, you can undo it right from the chat conversation.
- Clean room mode for sensitive dataIn clean room mode, only read-only tools run and contact emails and phone numbers are masked. It is designed for sensitive data, though anything you type directly is still sent to the approved provider.
- Choose your own AI modelThe assistant works with Claude, Gemini, DeepSeek, Cerebras, GLM and MiniMax, as well as OpenAI-compatible services, so your company is never locked into a single AI vendor.
- Personal API keys take priorityYour company stores AI provider API keys in its system settings. If a user configures their own personal API key, it takes precedence over the company key.
03How it works
How it works step by step.
How the assistant works
Open the assistant as a right-hand side panel on any page, or on the dedicated /chat page. Prompts are processed in the background in a dedicated queue and responses stream in real time. Conversations are saved and can be archived or restored at any time.
Read in detail
As it works, the assistant calls built-in tools: it searches and counts records, reads details and relationships, opens a page or filter, launches the mass editor, renders charts and creates direct links to records. Because it builds UI cards exclusively from predefined components, it can never hallucinate a non-existent page or broken link.
The tools are generic across all entities. Any new field, entity or action works with the assistant immediately—enforcing the exact same permission checks, validations and audit logs as standard forms and the REST API.
Changes happen only after you confirm
When you ask the assistant to create, update or delete a record, it writes nothing to the database by default. Instead, it prepares a draft and displays a “before and after” comparison card. The change is only executed once you confirm it, running as a standard save complete with validations, business rules and audit history. If required fields are missing, the assistant asks for them first. Dismissing the card leaves everything untouched.
Read in detail
Confirming the card re-runs the actual operation under your user account, so even an edited draft can never exceed your permissions. For bulk proposals, you can edit, merge, reorder or remove individual items before confirming.
Mass updates, record actions, relationship changes and menu edits are written immediately, always under your permissions. For mass updates, the assistant is instructed to verify the change and record count with you beforehand, though the system itself does not enforce a hard block. While a model can bypass draft review on an individual tool call, proposing a draft for approval is the default.
Clean room mode
For sensitive data, the assistant offers a clean room mode that you can select when starting a conversation. Only read-only tools approved for this mode are allowed to run. Contact emails and phone numbers are masked, file analysis returns only column headers and row counts, neither the user's identity nor page content is sent in the prompt, and sending emails is blocked.
Read in detail
Administrators control which modes users can start, can enforce clean room mode across all conversations, and can restrict which AI providers may be used. Governed execution rules can also enforce clean room mode for specific modules or entities; these rules only ever restrict access and never grant extra permissions. Clean room mode minimizes data exposure rather than creating an air-gapped environment: anything a user types directly into the chat is still sent to the approved provider.
Choosing AI models and API keys
The assistant supports Claude, Gemini, DeepSeek, Cerebras, GLM, MiniMax and OpenAI-compatible endpoints. The active model is selected in order of precedence: conversation setting, user preference, and company default, with automatic failover to a backup provider if the primary one is unavailable. API keys are managed in company settings, and a user's personal API key takes precedence when provided.
What the assistant can do across modules
Individual modules extend the assistant with domain-specific overviews. In CRM, it summarizes open opportunities, opportunities without a contact, leads without a follow-up activity and recent account activities, and proposes tasks directly from meeting notes. It can also display quote summaries, bank account balances and marketing campaign status.
Read in detail
In the warehouse, it summarizes product stock levels, low-stock items, today's stock movements, open stock transfers, and reserved but unissued goods. In manufacturing, it shows pending operations, open manufacturing and work orders, and reported labor by work center. It can also post messages to Mattermost and inspect or roll back data import runs.
04What runs automatically
Automations that save time.
Responses stream in real time
Once you send a prompt, the response is generated in the background in a dedicated ai-chat queue and streams onto your screen as it is written. You don't have to wait for the full answer to see what the assistant is writing.
No response is left hanging
Every minute, the system cleans up any interrupted responses. Conversations never get stuck on a message that looks like it is still generating, so you can keep working without interruption.
External integrations
- Claude as an assistant modelYou can select Claude as the AI assistant provider for a single conversation, a specific user or the entire company. Prompts are then processed by Claude using the API key configured in settings.
- Gemini as another optionGemini is another provider the assistant can use to answer prompts. If your primary provider is unavailable, a fallback provider takes over automatically so you can keep working.
- DeepSeek under admin controlDeepSeek is one of the supported AI assistant providers. Administrators can restrict which providers the assistant is allowed to use, keeping your company in control of where data is sent.
- Cerebras, even for a single chatThe assistant also works with Cerebras. Because you can set the model for an individual conversation, switching to it won't affect your user preferences or company-wide defaults.
- GLM with your company's API keyGLM is one of the supported model providers. You can store the API key in company settings, or let individual users enter a personal key that takes priority.
- MiniMax under the same permission rulesThe assistant can also run on MiniMax. Switching models never changes access control: the assistant always acts with the signed-in user's permissions and never sees anything they cannot access.
- Connect other AI servicesBeyond the built-in providers, the assistant supports any service with an OpenAI-compatible API, so your company can connect additional model endpoints that use this standard interface.
- Message teammates straight from chatThe assistant can send a message to Mattermost directly from your conversation, letting you pass information to colleagues without switching apps.
What the AI agent handles
Handles new fields immediately
The assistant includes general-purpose tools for reading, searching, counting and inspecting the schema of any entity (record type). That means it can work with new fields or entities right away.
Proposal first, write second
When you ask to create, update or delete a record, the assistant prepares a proposal by default. Nothing is saved until you confirm it, and dismissing the card leaves your data untouched.
Bulk changes within your permissions
The assistant executes bulk edits, record actions and relationship updates under your permissions and writes them immediately. For bulk edits, it is instructed to confirm with you beforehand, though the system does not enforce this.
Results directly in the app
The assistant can navigate the application, apply filters, select records, render charts and generate direct links. You see the results on standard application screens, not just as text in a chat window.
A person approves whenever money, a customer, or an exception is involved.
05Frequently asked questions
Can the AI assistant modify ERP data on its own?
By default, creating, updating or deleting a record only prepares a proposal that you must confirm. Bulk edits, record actions and relationship changes are written immediately, always under your user account and permissions.
Can the assistant see data I don’t have access to?
No. Every tool runs as the signed-in user through the exact same permission checks as the REST API and UI forms.
Which AI models does the assistant use?
It supports Claude, Gemini, DeepSeek, Cerebras, GLM, MiniMax and OpenAI-compatible services. The company sets the default model, and individual users can choose their own.
What is clean-room mode?
It is a conversation mode for sensitive data: read-only tools only, masked contact email addresses and phone numbers, and optionally only approved AI providers. An administrator can enforce it across all conversations.
- MCP for external AI agentsConnect Claude Code, Codex and other agents to your ERP under specific user permissions
- CRM and sales pipelineProspects, leads and sales opportunities with controlled stages, forecasting and cashflow
- Warehouse managementWarehouses, bin locations, goods receipts, issues, transfers and stocktakes built on an immutable movement ledger
- Production managementProduction and work orders, Gantt scheduling, labor tracking, and material issue
- Data import and migrationControlled import from CSV, Excel or API with preview, error logging and rollback
- AI document data extractionTurns invoices from emails or PDFs into draft documents ready for your review
Let’s start together
Tell us how your business works.
We will show you what a system that takes over part of that work could look like.
Book a consultation