> For the complete documentation index, see [llms.txt](https://docs.borndigital.ai/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.borndigital.ai/digital-agent/conversation-flow/nodes-explained/ai-node/tools.md).

# Tools

Built-in tools of the AI node 2.0: Knowledge base, Web search, Redirect to human, Transfer to node and End process.

Tools are actions the model may take while it prepares its answer. The model decides on its own whether and when to call a tool, based on the tool description and your [Instructions](https://docs.borndigital.ai/digital-agent/conversation-flow/nodes-explained/ai-node/essential-tab.md). Add tools in **Essential → Tools → + Add** and click a tool chip to configure it.

One turn can contain several tool calls. The node runs them, gives the results back to the model and repeats this until the model answers — at most 10 rounds of tool calls per turn.

## Knowledge base

Searches one knowledge-base index of the current project and gives the best matching snippets to the model.

<figure><img src="https://4261467870-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2F6iQTvxgRZRwPS1NgIGEb%2Fuploads%2F43ZIgX8w2SMKaVt1N3fV%2Fai-node-tool-knowledge-base.png?alt=media&amp;token=a59329e2-dbfa-40c9-9017-600c59a0e6f6" alt="Knowledge base tool settings"><figcaption><p>Knowledge base tool settings</p></figcaption></figure>

| Setting                              | Meaning                                                                                                                                                       | Default |
| ------------------------------------ | ------------------------------------------------------------------------------------------------------------------------------------------------------------- | ------- |
| **Index name**                       | The index to search. Only indexes of the current project are offered; one index per tool.                                                                     | –       |
| **Number of snippets**               | How many snippets the search returns.                                                                                                                         | 5       |
| **Broader context for each snippet** | Adds the neighbouring snippets of every hit, so the model sees the surrounding text. Local indexes only; when off, one neighbour on each side is still added. | off     |
| **Number of adjacent snippets**      | How many neighbours on each side are added (shown when the option above is on).                                                                               | 2       |

After a knowledge-base search, the node fills the variables `kb_document_id`, `kb_document_name` and `kb_document_url` with the documents that were used. See [Response variable and routing](https://docs.borndigital.ai/digital-agent/conversation-flow/nodes-explained/ai-node/response-variable-and-routing.md).

{% hint style="info" %}
Need to search **several indexes at once**, use **hybrid (full-text + semantic) search**, filter documents by **labels**, choose the index **from a variable**, or always include a small reference index in full? Use the [Indexer MCP server](https://docs.borndigital.ai/digital-agent/conversation-flow/nodes-explained/ai-node/indexer-mcp-server.md) instead of this tool.
{% endhint %}

How to build and maintain indexes is described in [Knowledge base](https://docs.borndigital.ai/digital-agent/advanced-functions/knowledge-base.md).

## Web search

Lets the model search the internet and answer with current information. The search runs at the LLM provider.

<figure><img src="https://4261467870-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2F6iQTvxgRZRwPS1NgIGEb%2Fuploads%2FygzY24Cu1WgKw63zj430%2Fai-node-tool-web-search.png?alt=media&amp;token=e6d850fb-724c-4cb4-8e3b-619424a797a9" alt="Web search tool settings"><figcaption><p>Web search tool settings</p></figcaption></figure>

| Setting                 | Meaning                                                                                                                                                                                                 | Default      |
| ----------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ------------ |
| **Search context size** | How much content from the found pages is given to the model: High, Medium or Low. More context means better answers, higher cost and latency.                                                           | Medium       |
| **Country**             | Prefers results relevant to the selected country.                                                                                                                                                       | none         |
| **Websites to search**  | Limits the search to these domains or subdomains, for example `example.com` or `blog.example.com` (up to 100). Paths such as `example.com/path` are not supported; a full URL is reduced to its domain. | all websites |

{% hint style="warning" %}

* Web search works only with providers and models that support it (OpenAI models). GPT-5 nano does not support it.
* With reasoning models, set **Effort** to at least **Low** (Configuration tab).
* Tell the model in the Instructions when it should search the web and which sources it should prefer.
  {% endhint %}

The response variable contains a `web_search` object with `used_web_search`, `web_search_count` and `annotations` (the cited sources).

## Redirect to human, Transfer to node, End process

These three tools let the model decide where the conversation goes next. They are configured the same way:

* **Description** – tells the model when to use the tool. It is prefilled with a default text; adapt it to your rules (for example *"Use this tool when the customer asks for an operator or is upset twice in a row."*).
* One or more **targets**:
  * **Transfer to** – the node the conversation continues to.
  * **When** (Rationale) – when this target is the right one, for example *"invoice and payment questions"* for a node that transfers to the finance team. Variables are allowed.

<figure><img src="https://4261467870-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2F6iQTvxgRZRwPS1NgIGEb%2Fuploads%2F2fkIuEwpgzpXDiXTA1mK%2Fai-node-tool-transfer.png?alt=media&amp;token=742e55fa-3669-4a44-b8e7-10d89dd0645a" alt="Transfer to node with one target"><figcaption><p>Transfer to node with one target</p></figcaption></figure>

Use **Add** for another target; targets can be duplicated, deleted and reordered.

| Tool                  | Typical target                                                                                             |
| --------------------- | ---------------------------------------------------------------------------------------------------------- |
| **Redirect to human** | A MESSAGE node that tells the user they are being connected to an operator, followed by the REDIRECT node. |
| **Transfer to node**  | Another AI node specialised in one topic (a "specialist agent"), or any other part of the flow.            |
| **End process**       | A closing MESSAGE node followed by an END node.                                                            |

When the model uses one of these tools, the flow continues to the chosen node instead of **Next**, and the response variable contains the chosen node in `target_node`. The editor creates the routing conditions for you.

{% hint style="info" %}
The turn in which a transfer happens is not added to the AI conversation history.
{% endhint %}
