Dynamic tools

Dynamic tools let Anvia search a large tool catalog and send only the most relevant tool definitions to the model for each turn.

Explore dynamic tools

Page	Learn how to
Tool sets	Group and execute concrete tool implementations.
Tool index	Build a searchable catalog and attach it to an agent.
Embedding text	Improve selection with accurate capability language.
Static and dynamic	Keep essential tools visible while retrieving the long tail.
Safety	Filter exposure and enforce authorization at execution.
Checklist	Verify relevance, permissions, and catalog freshness.

Why retrieve tools

Static tools are included in every model turn. That is the right default for a small, stable set, but a large catalog increases prompt size and makes tool choice less focused.

Dynamic selection adds a retrieval step:

Current prompt
    ↓
Search embedded tool definitions
    ↓
Apply metadata filter and threshold
    ↓
Send selected definitions to the model
    ↓
Execute the chosen tool from the backing ToolSet

The model sees only selected definitions. The application still owns every concrete tool implementation.

Build a dynamic catalog

import { AgentBuilder } from '@anvia/core'
import { createToolIndex } from '@anvia/core/tool'
import { vectorFilter } from '@anvia/core/vector-store'

const toolIndex = await createToolIndex(
  embeddingModel,
  allSupportTools,
  {
    metadata(tool) {
      return {
        productArea: tool.name.startsWith('billing_')
          ? 'billing'
          : 'support',
      }
    },
  },
)

const agent = new AgentBuilder('billing-support', model)
  .dynamicTools(toolIndex, {
    topK: 6,
    threshold: 0.72,
    filter: vectorFilter.eq('productArea', 'billing'),
  })
  .build()

Anvia searches again before every turn. Tool results can change the current runtime prompt, so a later turn may receive a different set of definitions.

Use it for scale, not authorization

Dynamic retrieval narrows the model-facing action surface. It does not prove that the user may execute a selected tool. Every handler must still validate input, enforce user and tenant permissions, protect side effects, and record sensitive actions.