AI and Assistant
AI in LadVen OS helps prepare work replies, analyze context, and perform allowed actions in tasks, CRM, calendar, and communications. The assistant does not replace the responsible person: a user chooses the scenario, checks the result, and owns the decision.
Use AI where it saves time and reduces routine, but do not delegate management responsibility, work acceptance, or sensitive-data handling to it.
Where to Find It
The main entry point is AI Hub (/ai). Depending on your role, it can include a playground, AI connections, scenarios, assistant profiles, bots, settings, and access controls for replies and actions.
If a tab is not available, check portal permissions or ask an administrator.
Direct routes open a specific area: /ai/playground, /ai/connections, /ai/scenarios, /ai/bots, /ai/settings.
The default /ai screen may be an administrative sandbox for prompt and JSON checks rather than a finished end-user chat. This depends on rollout and permissions; do not treat technical fields or mock responses as a production result. Open the relevant tab and confirm that it is available to your role before using a workflow.
When to Use the Assistant
Use the assistant to draft a reply, shorten a long discussion, extract next steps, improve a task brief, prepare a checklist, create an allowed reminder or calendar action, or review text before sending.
Do not use the assistant as the only decision source for customers, contracts, payments, personal data, or legally important actions.
Profiles and Scenarios
An assistant profile defines where it is useful and what it is allowed to do. One profile may be limited to replies, another may read tasks or CRM, and another may perform controlled actions.
Before using a profile, check its process, data access, action permissions, review owner, and limits for files, images, or tool calls.
If a profile does not fit the process, do not bypass the restriction by manually copying private data into the prompt.
Permissions and actions
Assistant replies and assistant actions carry different levels of risk. A reply can help word some text, while an action can create a reminder, change a work object, or start the next step.
Grant action permissions only to the roles and scopes where the boundaries of responsibility are clear. If an action affects a customer, a deadline, a task, or CRM, the user must check the result after it runs.
Playground
Use the playground to test a request before enabling a scenario for real work. It can compare a direct request with a saved scenario, select a connection, pass text, JSON context, or message history, request text or JSON output, and review a streaming response.
Use only anonymized examples. Replace names, phone numbers, amounts, links, and document numbers with demo values. When a prompt works, save it as a scenario instead of asking employees to copy a long prompt manually.
Connections
A connection defines which AI channel runs a scenario. Check the name, enabled state, access mode, image support, model, and timeout.
Keys and technical URLs must not appear in public instructions, screenshots, or user messages. Mark temporary connections clearly and do not use them for customer-facing commitments.
If a connection or scenario editor shows technical error text instead of a clear localized explanation, do not copy or publish it. Record only the route, time, and a safe description of the action, then ask the portal owner; do not blindly retry saving, refreshing, or deactivating.
Scenarios
A scenario turns a repeatable prompt into a controlled work operation. Define its purpose, output type, system instruction, user prompt template, input and output limits, overflow behavior, and response schema when structured output is needed.
Before enabling a scenario, check that the result is clear to the employee, the prompt does not ask for excessive private or commercial data, limits do not remove important context, JSON output fits the target form, and the employee knows where to verify and record the result.
Test scenarios that affect tasks, CRM, calendar, or documents in a demo process first.
Bots and Assistant Profiles
Bots are for automated or chat-based scenarios where the action should not run as a personal employee account. A bot actor needs a clear name, limited scenario, and activity only for the intended process.
An assistant profile connects a scenario, connection, and allowed tools. Check task read, CRM read, CRM write, image input, and the maximum number of tool calls per reply. The broader the profile, the stricter the human review should be.
AI in Work Surfaces
AI can appear outside AI Hub:
- in comments and chats to fix typos, make text businesslike, shorten, soften, or structure it;
- in tasks to suggest a title, draft a checklist, or shape readiness criteria;
- in CRM to parse text or a file for draft requisites;
- in the calendar to prepare a safe reminder or AI action.
Treat every result as a draft until an employee reviews it.
Security
Do not paste passwords, tokens, keys, payment data, unnecessary personal data, or internal technical URLs into prompts. If you need to describe a situation, anonymize the example and keep only the work-relevant meaning.
Before sending a reply prepared by AI, check the facts, tone, promises to the customer, deadlines, and participant names. The assistant can help with the text, but it must not invent commitments that do not exist.
Manager Control
A manager or administrator should regularly check which scenarios are enabled, which profiles can take actions, whether CRM write access is truly needed, whether vague scenarios like "do everything" exist, whether prompts include unnecessary customer data, and which AI drafts users often correct.
When AI is often wrong, adjust the scenario or permissions instead of teaching employees to ignore the risk.
Safe Demonstrations
When you show AI Hub to colleagues or prepare training materials, use demo connections, scenarios, prompts, and responses. The screen must not show real keys, base URLs, customer data, email addresses, phone numbers, contract amounts, customer names, or private files.
Good Practices
- Use AI for drafts, not final decisions without review.
- Create separate profiles for support, sales, tasks, and calendar work.
- Limit assistant actions when a hint or draft is enough.
- Check that the answer does not include private data from another customer or project.
- Record important decisions in a task, CRM record, or document, not only in the assistant reply.
- Disable scenarios that no longer match the process or current UI.
Detailed AI Hub pages
Related Scenarios