An AI tool is a skill or capability that the Thunk.AI platform gives an AI agent to extend what it can do, to bring in information it doesn't have, or for it to record its work. Instead of responding to an instruction with a message, an AI agent can respond by invoking an AI tool with appropriate parameters. You can think of a tool invocation as a structured response from the AI agent. Thunk.AI makes extensive use of tool invocations as the primary mechanism for AI agents to respond.
AI tools are used to connect an AI agent to the rest of the business work environment and the other applications and data in that environment. Some tools are used to fetch information (eg: search a company database) while other tools perform actions (eg: create new documents or update external systems).
Tools reach an AI agent in two ways:
Tool libraries that are ready to use: built into the platform, provided by a connection, or imported from another thunk. You turn them on or off; you do not write them.
Custom-built tools that you create for one thunk, from Add New Tool in the thunk's Custom-built tools pane.
Tool Libraries
Most designers of thunks do not have to build AI tools from scratch. Many AI tools come ready to use in bundles called AI Tool Libraries. The thunk's Connections pane lists them. There are different kinds of libraries:
Standard libraries: a default set of libraries available to every thunk.
Web: search the web, maps, places, and shopping results, and answer questions from web content. Tools that read a web page or save a page or file from a URL are also here; they start turned off, so enable them when a step needs them.
Communications: send an email, reply to an email, and draft a calendar event. Sending a new email starts turned off.
Images: generate an image from a description.
Audio/Video: extract information from audio or video, and grab still images at given timestamps.
PDFs: fill a PDF form and view specific pages of a long PDF. See Working with documents and PDFs.
Document Templates: create a document from a template the AI agent chooses at run time. See Generate documents from templates.
Application libraries: every user can augment their account with connections to other systems. For example, a Google Drive or OneDrive connection adds a library that lets the AI agent read and write files, spreadsheets, and folders there. Another common use of connections is to integrate with an existing enterprise application through an MCP server or a REST API. Learn more about integration with applications.
Imported thunk libraries: users can define AI tools in one thunk and export them to share and reuse across other thunks (of their own or for access to other users as well). There are many benefits to such modular reuse, especially when teams of users are implementing many AI agent automation processes. Learn more about modular reuse of thunks.
Plug-in connections: a thunk can declare the tools it needs as an interface, and whoever uses the thunk plugs in a connection that provides them. See Reuse thunk interfaces and plug-in connections.
Platform libraries: some of the internal capabilities of the platform (eg: reflection, or workflow state update) are also represented as AI tools. While the user cannot directly author these tools, they can be configured in the same way as any other AI tool.
Each content folder in a thunk also gets a search tool automatically (for example, search_documents), so the AI agent can find relevant information in that folder. See Content Folders.
Custom-built Tools
When no library tool does exactly what a step needs, the thunk owner or an admin can build a tool for this thunk. Open the thunk's Custom-built tools pane and choose Add New Tool. Every type starts the same way: you describe what the tool should do, and a builder AI agent writes the tool with you — its name, description, inputs, and implementation. You can watch it work, refine it in plain language, and run the tool with test values using Try it!.
Tool type | What it does | Use it when | Learn more |
Custom AI Tool | Runs its own AI agent with its own instructions and tools each time it is called | The task needs judgment or language understanding — reading an invoice, classifying a photo, drafting a reply | |
Custom Code Tool | Runs a JavaScript function in a secure sandbox | The result is fully determined by the inputs — a rule, a calculation, parsing a data payload from another system | |
Document Template Tool | Fills one fixed Word or Google Doc template and returns the new document | The workflow always produces the same kind of letter, form, or report | |
Spreadsheet Lookup Tool | Returns matching rows from one spreadsheet, filtered by the columns you choose | The answer lives in reference data in a sheet — a price list, a routing table, an approved-vendor list | |
Custom Database Tool | Runs a SQL query against a connected database | The answer lives in your own database tables | |
Tool Wrapper | Calls an existing tool with some of its arguments fixed, behind a name and inputs of your own | A library or connection tool is almost right, and you want a narrower, repeatable version of it |
Which type should I pick?
Start from where the answer or the action already lives:
An existing tool already does the work (web search, a connection's tool), but the AI agent keeps passing the wrong options or needs the same setup every time → Tool Wrapper.
The data is in a spreadsheet → Spreadsheet Lookup Tool. The data is in a SQL database → Custom Database Tool.
The output is a document built from a known template → Document Template Tool.
The logic is a fixed rule or calculation → Custom Code Tool. Code gives the same answer for the same inputs every time, and because it does not call an AI model it is usually the fastest and cheapest option.
The task needs reading, interpretation, or judgment → Custom AI Tool. A Custom AI Tool can itself use other tools, so it is also the way to package a multi-step piece of AI logic that several workflow steps reuse.
You need to call a system that has an API → add it as a connection (MCP server or REST API) rather than a custom tool. Its tools then appear as an application library. See Connecting to External Systems and Tools.
Custom-built tools are ordinary tools once they are built: you can enable them per step, test them with automated tool tests (see Testing custom tools), and export them in a library for other thunks to import.
Tool Configuration
It is important that an AI agent gets the AI tools it needs, but it is also important that unnecessary tools are not provided. These can add cost, delay, and even cause unreliable agentic behavior.
AI tools can be configured at the level of the thunk as a whole. Entire tool libraries can be enabled or disabled for the thunk, and individual tools within each enabled library can be further disabled.
Within every AI instruction at the level of a workflow step, the tool libraries and tools enabled at the thunk level are automatically enabled. They can be further configured at this level to disable some of the libraries and tools.
Enabled tools can also be constrained via natural language instructions. For example, a web search tool can be constrained within one step with a constraint like this: "only search within site fda.gov". If you want that constraint to hold every time, wrap the tool instead — see How to build a Tool Wrapper.
On a workflow step, you can also limit when an enabled tool is offered based on the current work item's property values. See Conditional tool availability.
Some connected tools take a URL and download the file themselves. If that file only works inside Thunk.AI, open the tool's Settings and turn on Convert URLs to signed download URLs for the parameter. See Pass a file to a connected tool with a signed download URL.