Dynamic context

Dynamic context automatically retrieves relevant documents before every agent turn. It gives the model focused knowledge without placing the full corpus in every request.

Explore dynamic context

Page	Learn how to
Add context	Attach a prepared vector index to an agent.
Formatting	Turn search results into concise model-ready documents.
Filters and permissions	Scope retrieval with trusted application state.
Multiple indexes	Combine independent knowledge sources without flooding the prompt.

How it works

Current prompt → vector search → filter and rank → format documents → model turn

For each turn, Anvia searches every registered dynamic-context index with the current runtime prompt. Matching results are converted into documents and sent with that turn's model request.

Retrieval runs again after a tool call. A later turn can therefore receive different documents as the conversation and tool results change.

Add automatic retrieval

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

const agent = new AgentBuilder('docs-support', model)
  .instructions('Use retrieved documentation when it is relevant.')
  .dynamicContext(docsIndex, {
    topK: 5,
    threshold: 0.72,
    filter: vectorFilter.eq('published', true),
  })
  .build()

docsIndex must implement VectorSearchIndex. Prepare and populate the index outside the request path; see Knowledges for ingestion, embeddings, and vector stores.

Choose the right source

Requirement	Use
Small facts safe for every run	Static .context(...)
Relevant documents selected every turn	.dynamicContext(...)
Optional, model-directed search	index.asTool(...)
Live or permissioned product data	A scoped tool
A large searchable tool catalog	.dynamicTools(...)

Dynamic context is read-only evidence. It should not replace service calls for current account state or actions that need validation, authorization, and audit.