Knowledge search and skills in chat

When a Mind searches the knowledge it was trained on, the answer shows a search step with the passages it found. When a step only loads one of the platform's skills, it appears as a chip instead of a full card. This page covers both, and how to start a chat with a skill already loaded.

Follow a knowledge search step

You do not have to ask for a search: ask your question as usual, and the Mind searches its training cards by itself when an answer needs them. The search appears in the answer as its own card. Click the card to open the step, or point at it and click Maximize. On a phone, tap the card.

The card and the open view report the search:

• The card counts the passages it found, such as 3 passages, or shows 0 passages when nothing matched.
• While the search runs, the open view reads Searching knowledge.
• The open view names the query the Mind searched with, or reads Knowledge search when there is no query text.
• When the search names a folder, the card shows the folder path before the query on desktop. A search narrowed to a folder reads that folder and its subfolders only.
• When the search failed, the card shows no count; open the step to read the error (the header reads Search similar when there is no query text).
• Copy results copies the results from the open view.
• When nothing came back, the open view reads No passages returned.

When a search finds nothing

The search reports why it came back empty, and each reason has its own message:

• Nothing matched the question closely enough: No relevant passages found in the knowledge base.
• The search was narrowed to a folder that holds no searchable cards: No vectorized cards found under folder "<folder>".
• The Mind has no Vectorized training cards at all: No vectorized knowledge is available for this mind.

If searches keep coming back empty, add cards on the Mind's Training Hub tab (Train a Mind with training cards), and review Similar results limit and Minimum similarity score in AI Config (Choose models and AI behaviour).

Read a passage

Every result in the open view is a passage card.

• A numbered chip shows the passage's position in the results.
• The match score appears as a percentage with one decimal place, for example 87.5%.
• The card's name is the training card the passage came from, or Untitled card when that card has no name.
• The passage's folder and tags appear as small labels when it has them.
• Chunk 2 / 5 shows which part of the card's text the passage covers.
• Show excerpt opens the passage text on the card; Hide excerpt closes it again.

Open a passage in the Training Hub

1. Open the knowledge search step and find the passage you want.
2. Click the passage card, or press Enter or Space on it. The action is labelled Open this training card in Hub.
3. The search view closes and the Mind opens on its Training Hub tab, with the card selected.

A passage opens only when it carries a reference to its card and to its Mind. Without one, the card is not clickable, and you can still read the passage here.

Start a conversation with a skill

A skill is usually loaded by the Mind when a task needs it, but you can start a chat with one already loaded:

1. Type / in the message box. The menu lists the skills and commands that fit the chat you are in.
2. Choose the skill's command, or type its name after the slash, for example /star-schema in a Warehouse Builder chat.
3. The skill joins your message as a pill. Send the message; the conversation starts with the skill loaded.

Each chat has its own catalogue of skills, and only the skills that fit the chat appear in the menu. The commands each chat offers are listed on Chat commands, shortcuts and privacy. A skill that cannot start from a slash command says so: the skill "<name>" cannot be started with a slash command.

Skills the Mind used

A skill is a DataMind playbook the Mind loads: one that ships with DataMind OS, one your organisation publishes, one shared with the Mind, or one written for it. Its chip carries the skill's title and version, and an earlier load in the same turn is shown in a muted style.

• The first load reads Using: {title} v{version}.
• A skill that was already loaded in this turn reads {title} v{version} already loaded.
• The chip's tooltip reads "A DataMind playbook shipped with this platform version" and repeats the skill's title and version.

Click the chip to open the step's details. The view shows the skill's title and version, and these rows:

Expand Skill text to read the skill's text. Copy text copies it, and the panel shows how many characters it has. If the skill was updated after this conversation loaded it, the panel warns: "Updated to v{version} since this conversation loaded v{loadedVersion}; the model follows this text from its next request." While the text loads, the panel reads Loading the skill text; if it cannot load, it reads "The skill text could not be loaded."

What a knowledge search looks through

• A search looks only through Vectorized training cards. Persistent cards are part of the Mind's instructions for every answer, so they do not appear as search results. See Train a Mind with training cards.
• The search's limits live on the Mind in AI Config: Similar results limit sets how many passages a search can return, and Minimum similarity score sets how close a match must be. See Choose models and AI behaviour.

Questions, answered

What does the percentage on a passage mean?

It is the match score between the search and the passage. The closer to 100, the stronger the match, and it is shown with one decimal place.

What does "Chunk 2 / 5" mean?

The training card's text is split into chunks for searching, and this passage covers chunk 2 of its 5 chunks.

Why does a passage say "Untitled card"?

The training card it came from has no name yet. Give the card a name in the Training Hub, if you can edit it.

Why does nothing happen when I click a passage?

Opening in the Hub needs a reference to the card and to its Mind. When the result does not carry one, the passage is not clickable, and you can only read its excerpt.

What is a skill?

A skill is a DataMind playbook the Mind loads, from the product, from your organisation, shared between Minds or written for this Mind. A step that loads one shows a chip with the skill's title and version. See Skills a Mind can load.

Why does a skill chip say "already loaded"?

The same skill was already loaded in this turn, so nothing was added again. The chip is shown in a muted style, and the step's State row reads Already in this turn, nothing added again.

Can I see the full text of a skill?

Yes. Open the skill's details and expand Skill text; Copy text copies the text, and the panel shows its length in characters.

Why does the skill text warn that it was updated?

The platform's copy of the skill changed after this conversation loaded it. The warning reads "Updated to v{version} since this conversation loaded v{loadedVersion}; the model follows this text from its next request."

Do I have to ask for a knowledge search?

No. The Mind searches its Vectorized training cards by itself when an answer needs them, and the search appears as a step in the answer.

Why did a knowledge search come back empty?

The card shows 0 passages, and the result says why: No relevant passages found in the knowledge base. means nothing matched closely enough, No vectorized cards found under folder "<folder>". means the folder holds no searchable cards, and No vectorized knowledge is available for this mind. means the Mind has no Vectorized training cards yet. Rephrase the question, check the folder, or add cards in the Mind's Training Hub, and review its similarity settings in Choose models and AI behaviour.

How do I start a chat with a skill already loaded?

Type / in the message box and choose the skill, or type its name after the slash, for example /star-schema. The skill joins your message as a pill, and sending the message starts the conversation with the skill loaded.

Why can't a skill be started with a slash command?

That skill is not offered for slash commands in this chat. Each chat has its own catalogue, and the / menu lists only the skills that fit it. The chat answers the skill "<name>" cannot be started with a slash command. Start the skill from a chat that offers it, or leave it to the Mind, which loads skills by itself when a task needs them.