What tools actually do
A tool is a concrete action. Fetch a webpage. Search a knowledge base. Generate a PDF. Send an email. Inside a network, a tool sits in a Tool step. Inside an agent, a tool is part of what the agent can use mid-conversation.
The tools described here are the same ones available to both networks and agents. Once you understand the catalog, you understand the action vocabulary of the whole platform.
You don't have to learn every tool to be productive. Most teams use five or six tools regularly. Skim this page once, then come back when you need something specific.
How tools are organized
Tools group into six families. Each family solves a different kind of problem.
Web research
Read the internet on your behalf — search engines, page contents, sites you care about.
Document & media generation
Produce something — a PDF, a spreadsheet, an image, an HTML mockup, a chart.
Code analysis
Read a repo, summarize a pull request, check dependencies, scan for security issues.
Knowledge search
Search inside the systems your team already lives in — Confluence, Jira, GitHub, custom indexes.
Utilities
The smaller building blocks — analyze a CSV, parse a regex, run OCR on a scanned document, transform JSON or YAML.
Communication
Send something out — an email today, with more channels arriving over time.
Web research
Web search
Run a search query and get back a list of results. Useful as the first step in any “go find out what’s happening with X” workflow.
Web crawler
Fetch the contents of a specific URL — or follow links a few pages deep. Useful when you know which site to read and want the text, not the search-engine summary.
A common pattern
A competitive-intelligence network uses Web search to find the latest articles about a competitor, then Web crawler to pull the full contents of the top three results, then hands those contents to an LLM Agent step for synthesis.
Document and media generation
These tools produce artifacts your team can share.
PDF generator
Produce a PDF from structured input. Useful for briefs, summaries, customer-facing documents, anything that needs to live as a file.
Document generator
Produce a Word-style document — DOCX. Useful when downstream tools expect editable documents.
Spreadsheet generator
Produce an XLSX. Useful for structured outputs — tables, summaries, breakdowns.
Image generator
Produce an image from a description. Useful for illustrations, concept art, social-media imagery.
Chart generator
Produce a chart from data — bar, line, pie, scatter, and more. Useful for adding visual summaries to a report.
HTML mockup generator
Produce a clickable HTML mockup of a layout or design idea. Useful for early product thinking, marketing pages, and “what would this look like?” prototypes.
QR code generator
Produce a QR code from a URL or short text. Useful for events, prints, links.
A common pattern
A monthly-review network uses Chart generator to produce three charts from the month’s data, Document generator to write a Word-style summary, and PDF generator to bundle them into a final deliverable.
Code analysis
These tools read software. Useful for engineering, product engineering, security review, and anything where understanding code is part of the workflow.
GitHub snapshot
Read a snapshot of a repo — files, structure, basic metrics. Useful as the first step when an agent needs to understand a codebase.
GitHub diff
Read the changes in a pull request or between two commits. Useful for PR review, changelog generation, and “what’s actually changed since last time?”
Dependency analysis
Read what a project depends on — packages, versions, licenses. Useful for compliance, supply-chain checks, and risk reviews.
Security scanning
Read code for known security risks. Useful as a checkpoint in a code-review network or a release-readiness review.
Pull-request review
Produce a structured review of a PR — what changed, what to look at, what to test, what looks risky. Useful as a first-pass review that human reviewers refine.
A common pattern
A “release readiness” network uses GitHub diff to read what’s changed, Dependency analysis to confirm nothing new requires licensing review, Security scanning to flag known vulnerabilities, and a final LLM Agent step to produce a release note.
Knowledge search
Search inside the systems your team already uses. Each search returns ranked, scoped results with citations back to the source.
Confluence search
Search a Confluence space — by meaning, not keyword. Useful for “what does our team say about X?”
Jira search
Search Jira projects. Useful for finding related work, prior decisions, similar tickets.
GitHub vector search
Search a GitHub repo by meaning — file contents, issue text, PR descriptions.
Federated search
Search across several sources at once. Useful when the answer might live in any of three or four systems.
Custom index search
Search a vector index your team built in VDF AI Data. Useful when you’ve curated a focused search surface for a specific use case.
A common pattern
A customer-support network uses Federated search to look across Confluence, Jira, and a custom index of past resolved tickets — and hands the top results to an agent that drafts a candidate reply.
Utilities
The smaller building blocks. Useful as middle steps that prepare data for the bigger ones.
CSV analysis
Read a CSV and produce a structured analysis — column types, ranges, distributions, anomalies.
JSON / YAML transform
Convert between formats, restructure, extract specific fields.
Regex tools
Apply pattern matching to extract or validate. Useful when an input has a known shape.
OCR
Read text from a scanned PDF or image. Useful when source documents aren’t digital text yet.
Sentiment analysis
Read text and classify the tone — positive, negative, neutral, with a confidence indicator. Useful for processing customer feedback.
A common pattern
A customer-feedback network uses CSV analysis to read the past week’s survey responses, Sentiment analysis to classify each one, and an LLM Agent step to produce a weekly themes summary.
Communication
Email send
Send an email. The most common end-of-network tool — deliver the result to the person who needs it.
A typical pattern is to combine the email tool with the document tools: generate a PDF, attach it, send the email.
A common pattern
A “compliance check” network ends with an Email send step that emails the legal team a link to the generated PDF, with the network’s summary in the body.
What’s not in the catalog (yet)
A few things teams ask about that don’t have dedicated tools yet — and what to do meanwhile:
| You want to… | Today’s path |
|---|---|
| Post to Slack | Use email to a Slack inbox, or a custom integration in your workspace |
| Update a row in a database | Use a custom integration; talk to your platform team |
| Trigger another network | Use an event trigger on the result of the first network |
The catalog grows over time. If your team uses something heavily that isn’t here yet, let us know — usage shapes the roadmap.
How tools are chosen at runtime
A tool only runs when a Tool step calls it. An agent can use a tool when its configuration allows it. In both cases:
- The instruction tells the tool what to do. The tool doesn’t guess.
- The output flows downstream. It becomes input for the next step or message.
- Errors are surfaced visibly. If a tool can’t complete, the step or the agent reports the problem rather than silently failing.
Tools are not magic. They do what they're told, with the inputs they're given. A good tool step has a clear instruction and clean inputs. A messy tool step produces messy output, no matter which tool it called.
A few patterns that work
Pair generation tools with knowledge search
The strongest documents come from a network that searches your team’s existing material first, then generates the new document on top of it. The Document generator on its own is a tool; the Document generator with Confluence search before it is a teammate.
End every triggered network with a communication step
If a network runs without a person watching, end it with an email or a delivery step. Otherwise the result lives only in the run history.
Use utilities to clean inputs
A Tool step that runs CSV analysis on a messy input before passing it to an LLM Agent often produces a noticeably better final result than the LLM Agent alone trying to make sense of the mess.
Where to go next
- Node types — the block that hosts a tool inside a network.
- Triggers and schedules — start a network at the right moment.
- Tools and knowledge — give an agent access to these same tools.
- Building workflows — compose tools and steps into a finished network.