Skip to main content
Apragya AI
Docs · AI & Automation

Tools: what your agent can do

The full catalog of tools agents can call, how tool grants work, how each call is sandboxed and metered, and how to register a custom tool.

Tools are the verbs an agent has access to. Without tools, an agent is a chatbot - it can reason and respond, but it can't take action. With tools, it can read invoices, query a CRM, post to an ERP, send Slack messages, fetch files from S3, call your own internal APIs. Every tool call is individually metered and audited.

Tool categories

  • Built-in business tools - invoice.read, po.lookup, grn.lookup, erp.post, crm.lookup, hr.lookup, contract.read, kyc.verify, and ~80 more
  • Communication tools - slack.send, teams.send, email.send, sms.send
  • Storage and file tools - s3.read, s3.write, sharepoint.read, drive.read
  • Integration tools - one per registered connector (salesforce.query, sap.invoke, quickbooks.create, …)
  • Custom tools - tenant-defined, see below

Granting tools to an agent

On the Agent Builder > Tools tab, add the specific tools the agent needs. The grant list IS the agent's permission set - the runtime blocks any tool call outside it. There's no "all tools" wildcard by design; over-grant is a real cost and security risk.

Tool calls write a RunNode row with status, input, output, cost, and latency. Open the agent's Runs tab to see a per-call breakdown for every run - useful for cost attribution and debugging slow runs.

Custom tools

  • Sidebar > Suites > Solution > Tools > New Tool
  • Define the tool's name, what inputs it takes, and where to send them - either a small in-app handler or one of your internal services
  • Give it a description so the agent's model knows when to reach for it
  • Set the AI Credits cost weight so the tool bills sensibly relative to a standard LLM call

Sandbox and dry-run modes

When you're iterating, point sensitive tools at staging endpoints or a dry-run mode so agent runs don't touch production systems. Configure this on the Test tab of the Agent Builder - overrides apply only to the tests you run there and never leak into live runs.

Need help building this? Talk to a real engineer

← All docs
Ask Vippy anything