NativeVectorSearchSkill

In-memory document search using TF-IDF-like word overlap scoring. No external dependencies or API keys required. Documents are provided via config and indexed at construction time.

Class: NativeVectorSearchSkill

Tools: search_documents

Env vars: None

Multi-instance: Yes

tool_name

string

Custom tool name for this instance. Required when using multiple instances.

documents

Record<string, unknown>[]

Array of documents to index. Each object has:

id (string, required) — Unique document identifier.
text (string, required) — Full text content of the document.
metadata (object, optional) — Metadata associated with the document.

num_results

numberDefaults to 3

Default number of top results to return.

distance_threshold

numberDefaults to 0

Minimum relevance score threshold.

1 import { AgentBase, NativeVectorSearchSkill } from '@signalwire/sdk';
2 
3 const agent = new AgentBase({ name: 'assistant', route: '/assistant' });
4 agent.setPromptText('You are a helpful assistant.');
5 
6 await agent.addSkill(new NativeVectorSearchSkill({
7   tool_name: 'search_faqs',
8   documents: [
9     { id: 'faq-1', text: 'To reset your password, visit the settings page...' },
10     { id: 'faq-2', text: 'Refund requests must be submitted within 30 days...' },
11   ],
12   num_results: 3,
13 }));
14 
15 agent.run();