This page is machine-translated from French. Read the French original.

Agent skills

A skill is a tool that an agent can call upon. They are activated for the entire instance by an administrator: there is no assignment per workspace.


Where to activate them

Settings → Agent → Agent Skills (Administrator) — Enables or disables a skill for the entire instance, and configures those that require it (access to a database, a search engine). A skill disabled here is not available anywhere; a skill enabled is available in all areas.

Space settings → Agent configuration — only sets the supplier and the agent model for this space.

Consequence: A sensitive skill activated for a specific use is activated for all spaces and for all accounts that can call @agent, including guests. Only the rights to messaging and calendar vary from person to person.


Catalog

Search and web

Live web search and navigation — queries a search engine and reports the results. Requires a search engine to be configured; several services are supported, including self-hosted engines for installations without external dependencies.

Retrieve websites — retrieves the content of a URL and extracts the relevant text. Works without configuration, but remains disabled until an administrator activates it.

These two capabilities send requests externally. On an isolated installation, they are irrelevant.

Documents and files

RAG and long-term memory — searches the documents in space and retains what the agent is asked to retain.

View and summarize documents — lists the documents in space and summarizes one of them based on its full text rather than extracts.

Access to the file system — lists, reads, writes, moves, and searches within an authorized directory. Each basic operation can be disabled separately—for example, allowing reading and disallowing writing. Available on the desktop application, not on a direct server installation.

Document creation — produces files (text, PDF, Word, Excel, PowerPoint) that the user downloads from the conversation.

Data

SQL Connector — connects to a database to list tables, read their structure, and execute queries. Supported database engines include PostgreSQL, MySQL, and SQL Server.

Create a dedicated read-only account. The agent constructs its own queries, and they execute without approval; nothing in WivenLLM restricts them to read access. Limited access to only the necessary tables is best practice and the only true protection.

Generate graphs — produces a graph from data collected during the task.

Communication

Outlook Connector and Gmail Connector — search for and read messages, prepare drafts, send and reply; Gmail also sorts messages (read, archive, trash). Sending, replying to, and deleting a draft require your approval each time, without an «Always allow» option; the same applies to deleting messages in Gmail. An agent session can send a maximum of five messages from Outlook.

An Outlook connection is read-only unless an administrator allows drafts and sending.

Google Calendar Connector — consults calendars and events, creates or modifies an event, responds to an invitation.

Microsoft Teams — searches for a person in the company directory, checks their Teams presence, and creates an internal reminder ticket. This skill does not read or post to Teams discussions. Ticket creation occurs without approval.

Each member connects their own Outlook, Gmail, and Google Calendar; the agent only uses the inbox of the person asking the question. Microsoft Teams, on the other hand, is configured only once for the entire instance.

→ Clients and connections

Memory and knowledge

Searching for conversations — It lists, searches, and reads your past conversations, including those in your other spaces. It never sees those of another user.

Second brain — consult and enrich your personal knowledge base.

→ Second Brain

Skills library — load a job instruction sheet upon request.

→ Skills library

Creation

Generate images And Generate videos — produce a visual during a task. Require a configured provider.

→ Images, videos and voices

Meetings

Minutes — structures a transcript or notes into minutes: participants, subjects, decisions, actions. The result is saved as a document in the space.

Automation

Create scheduled tasks — allows the agent to create a recurring task from a conversation, after your approval. Available in single-user mode only.

→ Planned tasks

Interaction

Request for clarification — allows the agent to ask you a question instead of making assumptions when the instructions are ambiguous. They will wait for your answer before proceeding, for a maximum of two minutes. This feature is in beta and is disabled by default; it can be enabled in the skill settings (three questions per turn by default).

Job skills

SEO Audit And Prospecting — two complete business applications of the desktop application, accessible both in conversation and through their own interface.

→ Business agents

Customized skills

Additional skills can be added by an administrator. These are programs: only install those whose origin you know.


Skills active by default

Three skills are activated by default because they remain within the system: RAG and long-term memory, viewing and summarizing documents, and searching for conversations. Second Brain is added when the function is activated. An administrator can deactivate them.

Everything else is activated on demand, including Website Recovery, SEO Audit and Prospecting, which reach the network.


Intelligent skills selection

The more tools an agent has, the more space the description of those tools takes up in its context — to the detriment of the task itself.

WivenLLM therefore only presents the agent with the skills relevant to the current request, rather than the entire set. The agent retains access to all the skills; they simply don't see them all on every turn.

The benefit is twofold: less context consumed, and better tool choices — an agent who sees five relevant tools makes fewer mistakes than an agent who sees fifty.

This setting, «Intelligence-based skill selection», is active by default and retains 15 tools. It can be modified in the skill settings.


Choose what to activate

Principle: the bare minimum. Each additional skill expands what an agent can do—even by mistake. The list applies to the entire instance: think about the most sensitive use case.

Use Reasonable skills
Documentary questions RAG and memory, document summary, conversation search
Monitoring and research Web search, site retrieval, summary
Data analysis SQL with a read-only account, graphs, document creation
Administrative assistance Calendar, messaging, document creation
Open to guests No sensitive skills

Questions to ask yourself before activating:

  • Who has access to this instance?
  • What can happen if the agent misuses this skill?
  • Does this capability send data externally?
  • Is there any human validation before the final effect? There is none for SQL queries, MCP tools, and agent flows.