Dynamic Tools for AI Agents
Providing tools to AI agents dynamically based on the current conversation context.
An interactive version of this guide is available in Membrane Console.
This guide uses Membrane API. Learn more here: Getting Started guide
Overview
Use this when you can't know the right tools ahead of time or when the tool catalog is very large.
✅ When to use:
- You manage many integrations and actions
- Tools depend on user intent or runtime context
- Users may connect new apps during the conversation
- You want to expose 'everything' without overwhelming the LLM
❌ When not to use:
- The toolset is small and predictable
- You want maximum caching efficiency
- You don't want extra infrastructure (search, indexing)
Step 1: Search actions tool
Give your agent a tool to search for available Actions using semantic search based on intent.
The agent can discover relevant tools dynamically during the conversation without having a pre-defined list.
searchActions
Step 2: Run action tool
Give your agent a tool to run the Action it selected from the search results. Pass the action ID and inputs provided by the agent, and return the output.
runAction
Run an action and return the result
// In your agent's tool handler
async function handleRunAction(actionId, input) {
const result = await membrane.action(actionId).run(input)
return result.output
}Updated 2 days ago