Manual: chat, agents and skills

How to ask the AI, attach files, run agents, approve proposals and create skills and agents.

First of all, an administrator or an AI manager needs to activate a provider in Configurações → IA → LLM (Settings → AI → LLM). Without it, the chat and the agents don’t answer.

The chat

Open Chat in the sidebar. Every conversation belongs to a project (collection): that’s where the AI gets its context, and that’s how conversations are organized for you to resume later. Administrators can also chat about the whole wiki.

Inside a document, the Chat com IA (Chat with AI) button, at the top, opens the same conversation in the side panel, with the document already as context. The ⌘. shortcut (Ctrl+. on Windows and Linux) opens or closes the last side panel used — the AI one by default.

  • Sources: the answer lists the documents it came from. Click to open.
  • Attachments: with the paperclip, send up to 3 files per message. From PDF, Word (.docx), text (.txt) and Markdown the AI reads the text; from images (PNG, JPG, WebP and GIF, up to 5 MB) it sees the content. A scanned PDF has no text to read, and there is no OCR: the AI says so instead of making things up.
  • @: in the message field, @ arms a skill or tells the AI to read a specific document. The choice becomes a chip above the field.
  • Stop: the stop button interrupts the answer; what already arrived stays on screen.
  • Rate: 👍 or 👎 on each answer. With 👎 you can give the reason — that’s what shows the team where the AI goes wrong.

Long conversations aren’t lost: when the history exceeds what the model can hold, the older parts become a summary for the AI, and you still see every message.

The AI drafting a document

Ask “write the minutes of this meeting” or “draft the specification from this card”. The chat shows the finished document, and you decide whether it goes into the wiki:

  • In Chat: Ver prévia (Preview) and Criar documento (Create document) — or Criar · {modelo} (Create · {template}).
  • In a document’s panel: Ver mudanças (See changes), Atualizar este (Update this one), Criar novo (Create new) or Descartar (Discard).

If the project has more than one template for that type and none was chosen, the chat asks which to use — the choice applies only to that document.

Agents

An agent is an AI with tools. While the chat answers with what is already written, the agent works: it researches the base, reads documents and attachments, applies skills and writes. If whoever configured the agent enabled those tools, it also queries ClickUp (with the token of whoever started it, in Configurações → Integrações — Integrations), searches the web, asks questions mid-task and calls other agents.

Running an agent

  1. In the chat, pick the agent in the project’s agent selector.
  2. Describe the task and send it. Some agents come with shortcuts for ready-made tasks.
  3. The agent runs in the background: you can leave the screen, and progress is saved in the conversation.
  4. If information is missing and it is allowed to ask, it asks, sometimes with options to choose from, and waits for your answer (up to 30 minutes, by default).

Each person runs one agent at a time in each project. To stop it midway, use Parar a execução (Stop the run).

What the agent saves and what it proposes

  • New document: the agent saves it right away, up to the creation limit configured for it. If it’s no good, move it to the trash.
  • Change to an existing document: becomes a proposal, which only goes into the wiki when you approve it. Use aprovar (approve) or the X to discard (with an optional reason). With several, Aprovar todas (N) (Approve all) asks for confirmation and applies them in list order.

Approving saves a new version of the document, which can be undone from the history.

Skills

A skill is a reusable instruction: “summarize in our minutes format”, “review the text in the brand’s tone”. Write it once and the whole team uses it in the chat with @. Agents can also use skills while working.

To create one: Configurações → IA → Skills. Give it a name, description and the instructions. A skill can apply to the whole workspace or to a single project, and can be tied to a document type — then it already knows the shape of what it produces.

Creating and tuning an agent

In Configurações → IA → Agentes (Agents) — administrator or AI manager; the manager only views the list:

FieldWhat it defines
Name, description and instructionsWho the agent is and how it works
Projeto dedicado (dedicated project, optional)Empty = shows up in every project and works in the open project. Filled = only in that one
ToolsWhat it can do. A new agent already comes with search, reading, edit proposals and document creation; you enable ClickUp, web search, asking and calling another agent
SkillsWhich skills it can use (empty = all)
Produced typeThe document type it drafts, with the project’s template
ModelIts own provider and model, if you don’t want the workspace default
Limites de execução (run limits)Model calls, documents created and tokens per run. Empty uses the system default

In each project, the Agentes section of Configurar projeto (Configure project) restricts which agents appear in that project’s selector. None checked = all that can appear there.

Document types and templates

In Configurações → Tipos de documento (Document types) you define the team’s templates: story, specification, test cases, minutes, and whatever else you need. Each type has a starting text and instructions for the AI. A project can use a variant of a type (for example, the specification in a client’s format) without changing the others’.

Conflict analysis

The AI compares the documents in a collection and points out the ones that contradict each other — a different deadline in two documents, a rule described two ways. There is no button in the wiki for this: run it with the MCP tool analisar_conflitos (see MCP server). With semantic search on, it only compares similar documents; without it, the AI sees a limited batch of the collection.

With semantic search turned on in the server (see Self-host), the chat and the agents find documents by the meaning of the question, and agents can check a statement against the base. ⌘K keeps searching by words.

In Configurações → IA → Busca semântica you choose the embedding model and, when turning it on or changing the model, use Reindexar tudo (Reindex everything) so existing content enters the index. Document types marked as non-indexable are left out.