Custom Tools
A custom tool is like teaching your agent a new skill from a manual you wrote — once it's connected, the agent can do that one job whenever it's useful. Each tool tells the agent how to call one outside service (look up an order, send a message, create a contact) and gives it back a tidy answer it can use in the conversation.
This section is the landing pad for everything about custom tools — how to make one with the AI helping you, how to wire one up by hand, and how to plug a finished tool into an agent.
What a custom tool is made of
A custom tool is just three small ingredients packaged together: a specific address that does one thing, an optional sign-in detail so the outside service trusts you, and a short description the agent reads to decide when to use it.
| Ingredient | What it means for you |
|---|---|
| The address Nirvai calls | The exact place on the outside service where the action happens (an "endpoint" — a specific address that does one thing, like the "send message" desk at a post office). |
| A sign-in detail | A saved credential (the sign-in details Nirvai keeps so it can talk to the service for you). Some services don't need one. |
| Parameters the agent fills in | The pieces of information the tool needs each time — like a customer ID or a search term. You decide which ones the agent fills in from the chat and which ones are always the same. |
| A name, description, and usage note | Plain-language labels so your agent knows what the tool is for and when to reach for it. |
Image: The Custom Tools page showing feature cards, stats, and the tool grid
What you can do here
| You can… | What that means for you |
|---|---|
| Let the AI build a set of tools for you | Pick a sign-in, describe what you want, and the AI reads the service's docs, sets everything up, tests it, and shows you the result for review. |
| Build a tool by hand | Fill in the address, the kind of request, and the fields yourself — best when you already know the service or it's a private one only you can see. |
| Decide what the agent fills in | Mark each field as "the agent fills it in" or "always the same value," so the agent has freedom where it should and guardrails where it shouldn't. |
| Test before saving | Try the tool with sample inputs and look at the real answer before you commit. |
| Give the tool to an agent | Assign finished tools to specific agents in the Agent Control Panel. |
Two ways to create a tool
AI Setup (recommended for most people). You pick a saved sign-in, say what you want the agent to be able to do, and the AI reads the service's documentation, sets up every tool, tests each one with a real call, and then shows you a tidy review screen — the Toolkit Guide — where you can look over every tool, see sample answers, and click Create when you're happy.
Manual. You fill in the address, the kind of request, and each field yourself. Best when you already know the service well, when the service is internal/private, or when you want every choice to be deliberate.
Giving tools to your agents
Tools live in your workspace once they're created. To actually let an agent use one, you assign it to that agent in the Agent Control Panel.
A few things to know:
- Each agent has room for 25 tools total. That single allowance is shared between custom tools and database tools — so if an agent already uses 10 database tools, it has room for 15 custom tools.
- One tool can be assigned to many agents. You don't have to rebuild it per agent.
- The agent only sees the name, description, usage note, and the fields you marked as "the agent fills it in." Everything else (the address, the sign-in, fixed values) stays behind the scenes.
What's next
Create tools with AI
Pick a sign-in, describe what you want, and let the AI research, set up, and test everything before you click Create.
Create a tool manually
Wire up one tool by hand with full control over the address, the kind of request, and every field.
Start from a template
Skip the blank page — start from a pre-filled setup for a popular service and tweak it.
Test your tool
Send sample inputs and inspect the real answer before saving or before handing the tool to an agent.
Use tools with agents
Assign finished tools to specific agents in the Agent Control Panel, within the shared 25-tool limit.