@agentforge/tools

Standard tools library with 88 production-ready tools.

Installation

pnpm add @agentforge/tools

Web Tools (22)

Web Search

import { webSearch } from '@agentforge/tools';

// Basic search (no API key needed - uses DuckDuckGo)
const result = await webSearch.invoke({
  query: 'TypeScript programming language',
  maxResults: 10
});

console.log(`Found ${result.results.length} results from ${result.source}`);
result.results.forEach(r => {
  console.log(`${r.title}: ${r.link}`);
  console.log(`  ${r.snippet}`);
});

// Premium search with Serper API (requires SERPER_API_KEY env var)
// Get your API key at: https://serper.dev
const premiumResult = await webSearch.invoke({
  query: 'Latest AI developments 2026',
  maxResults: 5,
  preferSerper: true  // Use Serper for Google search results
});

// Check metadata
console.log(`Source: ${premiumResult.source}`);
console.log(`Fallback used: ${premiumResult.metadata?.fallbackUsed}`);
console.log(`Response time: ${premiumResult.metadata?.responseTime}ms`);

Features:

• No API key required for basic searches (uses DuckDuckGo)
• Optional Serper API for premium Google search results
• Smart fallback: automatically switches providers when needed
• Returns structured results with titles, links, and snippets
• Tracks metadata (source, fallback usage, response time)

Environment Setup:

# Optional: Add to your .env file for premium Google search
SERPER_API_KEY=your-serper-api-key-here

HTTP Client

import { httpClient, httpGet, httpPost } from '@agentforge/tools';

// Generic HTTP request
const response = await httpClient.invoke({
  url: 'https://api.example.com/data',
  method: 'GET',
  headers: { 'Authorization': 'Bearer token' }
});

// GET request
const data = await httpGet.invoke({
  url: 'https://api.example.com/users'
});

// POST request
const result = await httpPost.invoke({
  url: 'https://api.example.com/users',
  body: { name: 'John' }
});

HTTP and scraper factories apply a destination policy before every request. By default, localhost, metadata, link-local, and private-network destinations are blocked, including destinations reached through redirects or hostname resolution. Use the explicit opt-in flags only for trusted operator-controlled or internal-network workflows:

import { createHttpTools, createScraperTools } from '@agentforge/tools';

const [safeHttpClient] = createHttpTools({
  destinationPolicy: { maxRedirects: 3 }
});

const [internalScraper] = createScraperTools({
  destinationPolicy: {
    allowPrivateNetwork: true,
    allowLocalhost: true,
  }
});

Available policy flags are allowLocalhost, allowPrivateNetwork, allowLinkLocal, allowMetadata, allowRedirects, and maxRedirects. See ST-11002 web egress policy hardening for the security rationale and migration guidance.

Web Scraping

import { webScraper, htmlParser, extractLinks } from '@agentforge/tools';

// Scrape webpage
const content = await webScraper.invoke({
  url: 'https://example.com',
  selector: '.content'
});

// Parse HTML
const parsed = await htmlParser.invoke({
  html: '<div>Content</div>',
  selector: 'div'
});

// Extract links
const links = await extractLinks.invoke({
  html: '<a href="/page">Link</a>'
});

URL Utilities

import { urlValidator, urlBuilder, urlQueryParser } from '@agentforge/tools';

// Validate URL - returns detailed URL components
const result = await urlValidator.invoke({
  url: 'https://example.com/path?query=value#hash'
});
console.log(result.data?.protocol);  // 'https:'
console.log(result.data?.hostname);  // 'example.com'
console.log(result.data?.pathname);  // '/path'

// Build URL from components
const url = await urlBuilder.invoke({
  protocol: 'https',
  hostname: 'api.example.com',
  pathname: '/users',
  query: { page: '1', limit: '10' }
});
console.log(url.url);  // 'https://api.example.com/users?page=1&limit=10'

// Parse query string - accepts URL or query string
const params = await urlQueryParser.invoke({
  input: 'https://example.com?foo=bar&baz=qux'
});
console.log(params.params);  // { foo: 'bar', baz: 'qux' }
console.log(params.count);   // 2

Slack Integration

import {
  sendSlackMessage,
  notifySlack,
  getSlackChannels,
  getSlackMessages,
  createSlackTools
} from '@agentforge/tools';

// Send a message to a Slack channel
const result = await sendSlackMessage.invoke({
  channel: 'general',  // Channel name (no #) or ID
  message: 'Hello from AgentForge! 👋'
});

// Send a notification with @mentions
const notification = await notifySlack.invoke({
  channel: 'alerts',  // Channel name (no #) or ID
  message: 'Deployment completed successfully!',
  mentions: ['john', 'jane']  // Usernames without @
});

// List available channels
const channels = await getSlackChannels.invoke({
  include_private: true  // Note: underscore, not camelCase
});

console.log(`Found ${channels.data?.count} channels`);
channels.data?.channels.forEach(ch => {
  console.log(`${ch.name} (${ch.num_members} members)`);
});

// Read message history
const messages = await getSlackMessages.invoke({
  channel: 'general',  // Channel name (no #) or ID
  limit: 10
});

console.log(`Retrieved ${messages.data?.count} messages`);
messages.data?.messages.forEach(msg => {
  console.log(`[${msg.timestamp}] ${msg.user}: ${msg.text}`);
});

// Custom configuration with factory function
const customSlackTools = createSlackTools({
  token: 'xoxb-your-custom-token'
});

// Use custom tools
const customResult = await customSlackTools.sendMessage.invoke({
  channel: 'custom',  // Channel name (no #) or ID
  message: 'Using custom token configuration'
});

Features:

• Send messages to any Slack channel
• Notify team members with @mentions
• List public and private channels
• Read message history with pagination
• Configurable via environment variables or programmatic configuration
• Full TypeScript support with Zod validation
• Structured logging for debugging

Environment Setup:

# Add to your .env file
# Use either a User Token or Bot Token
SLACK_USER_TOKEN=xoxp-your-user-token-here
# OR
SLACK_BOT_TOKEN=xoxb-your-bot-token-here

Getting a Slack Token:

1. Go to Slack API Apps
2. Create a new app or select an existing one
3. Navigate to "OAuth & Permissions"
4. Add required scopes:
   • User Token Scopes: channels:read, channels:history, chat:write, groups:read, groups:history
   • Bot Token Scopes: channels:read, channels:history, chat:write, groups:read, groups:history
5. Install the app to your workspace
6. Copy the token (starts with xoxp- for user tokens or xoxb- for bot tokens)

Programmatic Configuration:

// Override environment variables with custom configuration
const slackTools = createSlackTools({
  token: 'xoxb-your-custom-token'
});

// Use the custom tools (returns object with named properties)
const { sendMessage, notify, getChannels, getMessages } = slackTools;

Confluence Integration

import {
  searchConfluence,
  getConfluencePage,
  listConfluenceSpaces,
  getSpacePages,
  createConfluencePage,
  updateConfluencePage,
  archiveConfluencePage,
  createConfluenceTools
} from '@agentforge/tools';

// Search for pages across all spaces
const searchResult = await searchConfluence.invoke({
  query: 'API documentation',
  limit: 10
});

// Parse JSON string result
const searchData = JSON.parse(searchResult);
console.log(`Found ${searchData.count} pages (total: ${searchData.total})`);
searchData.results.forEach(page => {
  console.log(`${page.title} (${page.space})`);
  console.log(`  URL: ${page.url}`);
});

// Get a specific page by ID
const pageResult = await getConfluencePage.invoke({
  page_id: '123456'
});

// Parse JSON string result
const pageData = JSON.parse(pageResult);
console.log(`Title: ${pageData.page.title}`);
console.log(`Space: ${pageData.page.space}`);
console.log(`Content: ${pageData.page.content}`);
console.log(`Version: ${pageData.page.version}`);

// List all available spaces
const spacesResult = await listConfluenceSpaces.invoke({
  limit: 25
});

// Parse JSON string result
const spacesData = JSON.parse(spacesResult);
console.log(`Found ${spacesData.count} spaces`);
spacesData.spaces.forEach(space => {
  console.log(`${space.name} (${space.key})`);
});

// Get all pages in a specific space
const spacePagesResult = await getSpacePages.invoke({
  space_key: 'DOCS',
  limit: 50
});

// Parse JSON string result
const spacePagesData = JSON.parse(spacePagesResult);
console.log(`Found ${spacePagesData.count} pages in space ${spacePagesData.space}`);
spacePagesData.pages.forEach(page => {
  console.log(`${page.title} (ID: ${page.id})`);
});

// Create a new page
const newPageResult = await createConfluencePage.invoke({
  space_key: 'DOCS',
  title: 'New Documentation Page',
  content: '<p>This is the page content in Confluence storage format.</p>'
});

// Parse JSON string result
const newPageData = JSON.parse(newPageResult);
console.log(`Created page: ${newPageData.page.title}`);
console.log(`Page ID: ${newPageData.page.id}`);
console.log(`URL: ${newPageData.page.url}`);

// Create a child page
const childPageResult = await createConfluencePage.invoke({
  space_key: 'DOCS',
  title: 'Child Page',
  content: '<p>This is a child page.</p>',
  parent_page_id: '123456'
});

// Update an existing page
const updatedPageResult = await updateConfluencePage.invoke({
  page_id: '123456',
  title: 'Updated Title',
  content: '<p>Updated content.</p>'
});

// Parse JSON string result
const updatedPageData = JSON.parse(updatedPageResult);
console.log(`Updated page to version ${updatedPageData.page.version}`);

// Archive a page (move to trash)
const archivedPageResult = await archiveConfluencePage.invoke({
  page_id: '123456'
});

// Parse JSON string result
const archivedPageData = JSON.parse(archivedPageResult);
console.log(`Archived page: ${archivedPageData.page.title}`);

// Custom configuration with factory function
const customConfluenceTools = createConfluenceTools({
  apiKey: 'your-custom-api-key',
  email: 'custom@example.com',
  siteUrl: 'https://custom.atlassian.net'
});

// Use custom tools (returns object with named properties, not array)
const customResult = await customConfluenceTools.searchConfluence.invoke({
  query: 'search query',
  limit: 10
});

// Parse JSON string result
const customData = JSON.parse(customResult);
console.log(`Found ${customData.count} results`);

Features:

• Search pages across all Confluence spaces
• Retrieve full page content with metadata
• List all available spaces in your Confluence instance
• Get all pages within a specific space
• Create new pages with optional parent pages (hierarchical structure)
• Update existing page content and metadata
• Archive pages (move to trash)
• Configurable via environment variables or programmatic configuration
• Full TypeScript support with Zod validation
• Structured logging with [[tools:confluence]] prefix for debugging

Environment Setup:

# Add to your .env file
ATLASSIAN_API_KEY=your-atlassian-api-key-here
ATLASSIAN_EMAIL=your-email@example.com
ATLASSIAN_SITE_URL=https://your-site.atlassian.net

Getting Confluence API Credentials:

1. Go to Atlassian Account Settings
2. Click "Create API token"
3. Give it a descriptive label (e.g., "AgentForge Integration")
4. Copy the generated API token
5. Use your Atlassian account email and the API token for authentication
6. Your site URL is your Confluence instance URL (e.g., https://yourcompany.atlassian.net)

Programmatic Configuration:

// Override environment variables with custom configuration
const confluenceTools = createConfluenceTools({
  apiKey: 'your-custom-api-key',
  email: 'custom@example.com',
  siteUrl: 'https://custom.atlassian.net'
});

// Use the custom tools (returns object with named properties)
const { searchConfluence, getConfluencePage, listConfluenceSpaces, getSpacePages, createConfluencePage, updateConfluencePage, archiveConfluencePage } = confluenceTools;

Content Format: Confluence uses "storage format" (a subset of HTML) for page content. Common elements:

<p>Paragraph text</p>
<h1>Heading 1</h1>
<h2>Heading 2</h2>
<ul><li>Bullet point</li></ul>
<ol><li>Numbered list</li></ol>
<ac:structured-macro ac:name="code">
  <ac:plain-text-body><![CDATA[code block]]></ac:plain-text-body>
</ac:structured-macro>

Data Tools (31)

JSON Processing

import { 
  jsonParser, 
  jsonStringify, 
  jsonQuery,
  jsonValidator,
  jsonMerge 
} from '@agentforge/tools';

// Parse JSON
const parseResult = await jsonParser.invoke({
  json: '{"name": "John"}'
});
if (parseResult.success) {
  console.log(parseResult.data?.data); // { name: 'John' }
  console.log(parseResult.data?.type); // 'object'
}

// Stringify
const stringifyResult = await jsonStringify.invoke({
  data: { name: 'John' },
  pretty: true
});
if (stringifyResult.success) {
  console.log(stringifyResult.data?.json); // '{\n  "name": "John"\n}'
  console.log(stringifyResult.data?.length); // 24
}

// Query JSON (dot notation)
const queryResult = await jsonQuery.invoke({
  data: { users: [{ name: 'John' }] },
  path: 'users[0].name'
});
if (queryResult.success) {
  console.log(queryResult.data?.value); // 'John'
  console.log(queryResult.data?.type); // 'string'
}

// Validate JSON
const validationResult = await jsonValidator.invoke({
  json: '{"name": "John"}',
  schema: { type: 'object' }
});
if (validationResult.success) {
  console.log(validationResult.data?.valid); // true
  console.log(validationResult.data?.message); // 'Valid JSON'
}

// Merge JSON objects (uses .implement, returns raw value)
const merged = await jsonMerge.invoke({
  objects: [{ a: 1 }, { b: 2 }]
});
console.log(merged); // { a: 1, b: 2 }

CSV Processing

import { csvParser, csvGenerator, csvToJson, jsonToCsv } from '@agentforge/tools';

// Parse CSV
const parseResult = await csvParser.invoke({
  csv: 'name,age\nJohn,30',
  delimiter: ','
});
if (parseResult.success) {
  console.log(parseResult.data); // [{ name: 'John', age: '30' }]
  console.log(parseResult.rowCount); // 1
  console.log(parseResult.columnCount); // 2
}

// Generate CSV
const generateResult = await csvGenerator.invoke({
  data: [{ name: 'John', age: 30 }]
});
if (generateResult.success) {
  console.log(generateResult.csv); // 'name,age\nJohn,30\n'
  console.log(generateResult.rowCount); // 1
}

// Convert CSV to JSON
const csvToJsonResult = await csvToJson.invoke({
  csv: 'name,age\nJohn,30'
});
if (csvToJsonResult.success) {
  console.log(csvToJsonResult.json); // '[{"name":"John","age":"30"}]'
  console.log(csvToJsonResult.recordCount); // 1
}

// Convert JSON to CSV
const jsonToCsvResult = await jsonToCsv.invoke({
  json: '[{"name":"John","age":30}]'
});
if (jsonToCsvResult.success) {
  console.log(jsonToCsvResult.csv); // 'name,age\nJohn,30\n'
  console.log(jsonToCsvResult.rowCount); // 1
}

XML Processing

import { xmlParser, xmlGenerator, xmlToJson, jsonToXml } from '@agentforge/tools';

// Parse XML
const parseResult = await xmlParser.invoke({
  xml: '<root><name>John</name></root>'
});
if (parseResult.success) {
  console.log(parseResult.data); // { root: { name: 'John' } }
}

// Generate XML
const generateResult = await xmlGenerator.invoke({
  data: { root: { name: 'John' } }
});
if (generateResult.success) {
  console.log(generateResult.xml); // '<root><name>John</name></root>'
}

// Convert XML to JSON
const xmlToJsonResult = await xmlToJson.invoke({
  xml: '<root><name>John</name></root>',
  pretty: true
});
if (xmlToJsonResult.success) {
  console.log(xmlToJsonResult.json); // '{\n  "root": {\n    "name": "John"\n  }\n}'
}

// Convert JSON to XML
const jsonToXmlResult = await jsonToXml.invoke({
  json: '{"name":"John"}',
  rootName: 'person',
  format: true
});
if (jsonToXmlResult.success) {
  console.log(jsonToXmlResult.xml); // '<person>\n  <name>John</name>\n</person>'
}

Data Transformation

import { 
  arrayFilter, 
  arrayMap, 
  arraySort,
  arrayGroupBy,
  objectPick,
  objectOmit 
} from '@agentforge/tools';

// Filter array
const filterResult = await arrayFilter.invoke({
  array: [{ age: 25 }, { age: 30 }, { age: 35 }],
  property: 'age',
  operator: 'greater-than',
  value: 28
});
console.log(filterResult.filtered); // [{ age: 30 }, { age: 35 }]
console.log(filterResult.filteredCount); // 2

// Map array (extract properties)
const mapResult = await arrayMap.invoke({
  array: [{ name: 'John', age: 30 }, { name: 'Jane', age: 25 }],
  properties: ['name']
});
console.log(mapResult.mapped); // [{ name: 'John' }, { name: 'Jane' }]
console.log(mapResult.count); // 2

// Sort array
const sortResult = await arraySort.invoke({
  array: [{ score: 3 }, { score: 1 }, { score: 2 }],
  property: 'score',
  order: 'asc'
});
console.log(sortResult.sorted); // [{ score: 1 }, { score: 2 }, { score: 3 }]
console.log(sortResult.count); // 3

// Group by
const groupResult = await arrayGroupBy.invoke({
  array: [{ type: 'a', val: 1 }, { type: 'a', val: 2 }, { type: 'b', val: 3 }],
  property: 'type'
});
console.log(groupResult.groups); // { a: [{ type: 'a', val: 1 }, { type: 'a', val: 2 }], b: [{ type: 'b', val: 3 }] }
console.log(groupResult.groupCount); // 2
console.log(groupResult.totalItems); // 3

Neo4j Graph Database (7 tools)

Production-ready Neo4j integration with support for knowledge graphs, GraphRAG, and vector search.

Quick Start

# Install Neo4j driver (peer dependency)
pnpm add neo4j-driver

# Set environment variables
NEO4J_URI=bolt://localhost:7687
NEO4J_USER=neo4j
NEO4J_PASSWORD=your-password

# For embedding-enabled tools (optional)
EMBEDDING_PROVIDER=openai  # Options: openai, cohere, huggingface, voyage, ollama
OPENAI_API_KEY=sk-...      # Or appropriate key for your provider

Initialize Connection

import { initializeNeo4jTools } from '@agentforge/tools';

// Initialize from environment variables
await initializeNeo4jTools();

// Or with custom config
import { neo4jPool } from '@agentforge/tools';
await neo4jPool.initialize({
  uri: 'bolt://localhost:7687',
  username: 'neo4j',
  password: 'your-password',
  database: 'neo4j',
});

1. neo4jQuery - Execute Cypher Queries

import { neo4jQuery } from '@agentforge/tools';

// Create nodes and relationships
const result = await neo4jQuery.execute({
  cypher: `
    CREATE (p:Person {name: $name, age: $age})
    RETURN p
  `,
  parameters: { name: 'Alice', age: 30 }
});

console.log('Records:', result.data);
console.log('Summary:', result.summary);

2. neo4jGetSchema - Inspect Graph Schema

import { neo4jGetSchema } from '@agentforge/tools';

const schema = await neo4jGetSchema.execute({});

console.log('Node labels:', schema.schema.nodeLabels);
console.log('Relationship types:', schema.schema.relationshipTypes);
console.log('Property keys:', schema.schema.propertyKeys);
console.log('Indexes:', schema.schema.indexes);
console.log('Constraints:', schema.schema.constraints);

3. neo4jFindNodes - Find Nodes by Label/Properties

import { neo4jFindNodes } from '@agentforge/tools';

// Find all persons
const persons = await neo4jFindNodes.execute({
  label: 'Person',
  limit: 10
});

// Find with property filters
const adults = await neo4jFindNodes.execute({
  label: 'Person',
  properties: { age: 30 },
  limit: 5
});

console.log(`Found ${adults.count} nodes`);
adults.nodes.forEach(node => {
  console.log(node.properties);
});

4. neo4jTraverse - Graph Traversal

import { neo4jTraverse } from '@agentforge/tools';

// Traverse from a starting node
const result = await neo4jTraverse.execute({
  startNodeId: 123,
  relationshipType: 'KNOWS',  // Single relationship type (optional)
  direction: 'outgoing',       // lowercase: 'outgoing', 'incoming', or 'both'
  maxDepth: 2,
  limit: 20
});

console.log(`Found ${result.count} paths`);
result.paths.forEach(path => {
  console.log('Start node:', path.start);
  console.log('End node:', path.end);
  console.log('Relationships:', path.relationships);
  console.log('Depth:', path.depth);
});

5. neo4jVectorSearch - Semantic Search

import { neo4jVectorSearch } from '@agentforge/tools';

// Search with pre-computed embedding
const embedding = [0.1, 0.2, 0.3, /* ... 1536 dimensions */];

const results = await neo4jVectorSearch.execute({
  indexName: 'document_embeddings',
  queryVector: embedding,
  limit: 5
});

console.log(`Found ${results.count} similar nodes`);
results.results.forEach(item => {
  console.log(`Score: ${item.score}`);
  console.log('Node:', item.node.properties);
});

6. neo4jVectorSearchWithEmbedding - Semantic Search (Auto-Embedding)

import { neo4jVectorSearchWithEmbedding } from '@agentforge/tools';

// Search with plain text - embedding generated automatically!
const results = await neo4jVectorSearchWithEmbedding.execute({
  indexName: 'document_embeddings',
  queryText: 'How do graph databases work?',
  limit: 5
});

console.log('Query:', results.query.text);
console.log('Embedding model:', results.embedding.model);
console.log('Embedding dimensions:', results.embedding.dimensions);

results.results.forEach(item => {
  console.log(`${item.node.properties.title} (score: ${item.score})`);
});

7. neo4jCreateNodeWithEmbedding - Create Nodes with Auto-Embedding

import { neo4jCreateNodeWithEmbedding } from '@agentforge/tools';

// Create node with automatic embedding generation
const result = await neo4jCreateNodeWithEmbedding.execute({
  label: 'Document',
  properties: {
    title: 'Introduction to Neo4j',
    content: 'Neo4j is a powerful graph database...',
    category: 'database'
  },
  textProperty: 'content',        // Property to generate embedding from
  embeddingProperty: 'embedding'  // Where to store the embedding
});

console.log('Node created:', result.nodeId);
console.log('Embedding dimensions:', result.embedding.dimensions);
console.log('Model used:', result.embedding.model);

Embedding Providers

AgentForge supports 5 embedding providers for maximum flexibility:

Provider	Best For	Dimensions	Cost	Setup
OpenAI	Production, quality	1536-3072	Low	OPENAI_API_KEY
Cohere	Multilingual	384-1024	Medium	COHERE_API_KEY
HuggingFace	Open-source	384-1024	Low-Free	HUGGINGFACE_API_KEY
Voyage	Domain-specific	1024-1536	Medium	VOYAGE_API_KEY
Ollama	Privacy, offline	384-1024	Free	Local (no key)

Switch providers easily:

# OpenAI (default)
EMBEDDING_PROVIDER=openai
OPENAI_API_KEY=sk-...

# Cohere
EMBEDDING_PROVIDER=cohere
COHERE_API_KEY=...

# HuggingFace
EMBEDDING_PROVIDER=huggingface
HUGGINGFACE_API_KEY=...

# Voyage AI
EMBEDDING_PROVIDER=voyage
VOYAGE_API_KEY=...

# Ollama (local, no API key!)
EMBEDDING_PROVIDER=ollama
OLLAMA_BASE_URL=http://localhost:11434  # Optional

Ollama Setup (Privacy-Focused):

# Install Ollama
curl -fsSL https://ollama.ai/install.sh | sh

# Pull an embedding model
ollama pull nomic-embed-text

# Start Ollama
ollama serve

# Use in AgentForge (no API key needed!)
EMBEDDING_PROVIDER=ollama

GraphRAG Example

Complete GraphRAG implementation with automatic embeddings:

import {
  neo4jQuery,
  neo4jCreateNodeWithEmbedding,
  neo4jVectorSearchWithEmbedding
} from '@agentforge/tools';

// 1. Create vector index (run once)
await neo4jQuery.execute({
  cypher: `
    CREATE VECTOR INDEX document_embeddings IF NOT EXISTS
    FOR (n:Document)
    ON (n.embedding)
    OPTIONS {
      indexConfig: {
        \`vector.dimensions\`: 1536,
        \`vector.similarity_function\`: 'cosine'
      }
    }
  `
});

// 2. Add documents with automatic embeddings
const documents = [
  {
    title: 'Neo4j Guide',
    content: 'Neo4j is a graph database that stores data as nodes and relationships.'
  },
  {
    title: 'GraphRAG Overview',
    content: 'GraphRAG combines graph databases with retrieval-augmented generation.'
  },
  {
    title: 'Vector Search',
    content: 'Vector search enables semantic similarity matching using embeddings.'
  }
];

for (const doc of documents) {
  await neo4jCreateNodeWithEmbedding.execute({
    label: 'Document',
    properties: doc,
    textProperty: 'content',
    embeddingProperty: 'embedding'
  });
}

// 3. Query using semantic search
const results = await neo4jVectorSearchWithEmbedding.execute({
  indexName: 'document_embeddings',
  queryText: 'What is GraphRAG?',
  limit: 3
});

console.log(`Found ${results.count} relevant documents:`);
results.results.forEach(item => {
  console.log(`- ${item.node.properties.title} (score: ${item.score})`);
});

Key Benefits:

• No manual embedding generation
• Provider-agnostic (switch with env vars)
• Automatic dimension handling
• Built-in retry logic
• Production-ready error handling

Relational Database Tools (6)

Vendor-agnostic tools for PostgreSQL, MySQL, and SQLite. See the Database Tools Guide for concepts and the Database Agent Tutorial for a walkthrough.

Installation:

pnpm add @agentforge/tools pg        # PostgreSQL
pnpm add @agentforge/tools mysql2    # MySQL
pnpm add @agentforge/tools better-sqlite3  # SQLite

Return Pattern

All relational tools return { success: true, ... } on success or { success: false, error: string } on failure. They do not throw for database or query errors.

Exception: If the required vendor driver (pg, mysql2, or better-sqlite3) is not installed, the tool will throw a MissingPeerDependencyError synchronously. Ensure peer dependencies are installed to avoid this.

relationalQuery

Execute raw SQL with parameterized binding:

import { relationalQuery } from '@agentforge/tools';

const result = await relationalQuery.invoke({
  sql: 'SELECT id, email FROM users WHERE status = $1 LIMIT 10',
  params: ['active'],
  vendor: 'postgresql',
  connectionString: 'postgresql://user:pass@localhost:5432/mydb',
});

// Success: { success: true, rows: [...], rowCount: 10, executionTime: 12 }
// Error:   { success: false, error: '...', rows: [], rowCount: 0 }

Parameter	Type	Required	Description
sql	string	Yes	SQL query with placeholders ($1 for PG, ? for MySQL/SQLite)
params	unknown[] | Record<string, unknown>	No	Positional or named parameters
vendor	'postgresql' | 'mysql' | 'sqlite'	Yes	Database vendor
connectionString	string	Yes	Database connection string

relationalSelect

Type-safe SELECT without writing SQL:

import { relationalSelect } from '@agentforge/tools';

const result = await relationalSelect.invoke({
  table: 'orders',
  columns: ['id', 'total', 'status'],
  where: [
    { column: 'status', operator: 'eq', value: 'pending' },
    { column: 'total', operator: 'gt', value: 100 },
  ],
  orderBy: [{ column: 'total', direction: 'desc' }],
  limit: 10,
  offset: 0,
  vendor: 'postgresql',
  connectionString: DB_URL,
});

// Success: { success: true, rows: [...], rowCount: 10, executionTime: 8 }

Parameter	Type	Required	Description
table	string	Yes	Table name (e.g. users or public.users)
columns	string[]	No	Column filter (omit for SELECT *)
where	WhereCondition[]	No	WHERE conditions (combined with AND)
orderBy	OrderBy[]	No	ORDER BY clauses
limit	number	No	Max rows to return
offset	number	No	Rows to skip
streaming	StreamingOptions	No	Chunked streaming mode
vendor	string	Yes	Database vendor
connectionString	string	Yes	Connection string

WhereCondition:

{ column: string, operator: Operator, value?: ScalarValue }

Operators: eq, ne, gt, lt, gte, lte, like, in, notIn, isNull, isNotNull

StreamingOptions:

Field	Type	Default	Description
enabled	boolean	true	Enable chunked streaming
chunkSize	number	100	Rows per chunk (1–5000)
maxRows	number	—	Cap on total streamed rows
sampleSize	number	50	Rows included in response payload (0–5000)
benchmark	boolean	false	Memory benchmark mode

relationalInsert

Insert single rows or batch arrays:

import { relationalInsert } from '@agentforge/tools';

// Single row with RETURNING
const result = await relationalInsert.invoke({
  table: 'users',
  data: { email: 'alice@example.com', name: 'Alice' },
  returning: { mode: 'id', idColumn: 'id' },
  vendor: 'postgresql',
  connectionString: DB_URL,
});
// { success: true, rowCount: 1, insertedIds: [42], rows: [], executionTime: 5 }

// Batch insert
const batch = await relationalInsert.invoke({
  table: 'events',
  data: events, // Array of records
  batch: { batchSize: 200, continueOnError: true },
  vendor: 'postgresql',
  connectionString: DB_URL,
});
// { success: true, rowCount: 1000, batch: { successfulItems: 1000, ... } }

Parameter	Type	Required	Description
table	string	Yes	Table name
data	Record | Record[]	Yes	Row(s) to insert
returning	{ mode, idColumn? }	No	Return mode: 'none' (default), 'id', 'row'
batch	BatchOptions	No	Batch execution controls
vendor	string	Yes	Database vendor
connectionString	string	Yes	Connection string

BatchOptions (shared by Insert, Update, Delete):

Field	Type	Default	Description
enabled	boolean	true	Enable batched processing
batchSize	number	100	Items per chunk (1–5000)
continueOnError	boolean	true	Continue on chunk failure
maxRetries	number	0	Retries per failed chunk (0–5)
retryDelayMs	number	0	Delay between retries in ms (0–60000)
benchmark	boolean	false	Synthetic benchmark mode

relationalUpdate

Conditional updates with safety guards:

import { relationalUpdate } from '@agentforge/tools';

// Single update
const result = await relationalUpdate.invoke({
  table: 'users',
  data: { status: 'verified' },
  where: [{ column: 'id', operator: 'eq', value: 42 }],
  vendor: 'postgresql',
  connectionString: DB_URL,
});
// { success: true, rowCount: 1, executionTime: 3 }

// Batch update (multiple independent operations)
const batch = await relationalUpdate.invoke({
  table: 'products',
  operations: [
    { data: { price: 9.99 }, where: [{ column: 'id', operator: 'eq', value: 1 }] },
    { data: { price: 19.99 }, where: [{ column: 'id', operator: 'eq', value: 2 }] },
  ],
  batch: { batchSize: 50 },
  vendor: 'postgresql',
  connectionString: DB_URL,
});

Parameter	Type	Required	Description
table	string	Yes	Table name
data	Record	Conditional	Column-value pairs (single mode)
where	WhereCondition[]	Conditional	Required unless allowFullTableUpdate: true
allowFullTableUpdate	boolean	No	Explicit opt-in for no-WHERE update
optimisticLock	{ column, expectedValue }	No	Optimistic locking condition
operations	UpdateOperation[]	Conditional	Batch mode (mutually exclusive with data)
batch	BatchOptions	No	Batch execution controls

relationalDelete

Safe deletion with WHERE requirement:

import { relationalDelete } from '@agentforge/tools';

// Hard delete
const result = await relationalDelete.invoke({
  table: 'sessions',
  where: [{ column: 'expires_at', operator: 'lt', value: '2026-01-01' }],
  vendor: 'postgresql',
  connectionString: DB_URL,
});
// { success: true, rowCount: 150, softDeleted: false, executionTime: 12 }

// Soft delete
const soft = await relationalDelete.invoke({
  table: 'users',
  where: [{ column: 'id', operator: 'eq', value: 42 }],
  softDelete: { column: 'deleted_at' },
  vendor: 'postgresql',
  connectionString: DB_URL,
});
// { success: true, rowCount: 1, softDeleted: true }

Parameter	Type	Required	Description
table	string	Yes	Table name
where	WhereCondition[]	Conditional	Required unless allowFullTableDelete: true
allowFullTableDelete	boolean	No	Explicit opt-in for full-table delete
cascade	boolean	No	Cascade-aware error messaging
softDelete	{ column?, value? }	No	Soft-delete mode (column defaults to 'deleted_at')
operations	DeleteOperation[]	Conditional	Batch delete operations
batch	BatchOptions	No	Batch execution controls

relationalGetSchema

Introspect tables, columns, keys, and indexes:

import { relationalGetSchema } from '@agentforge/tools';

const result = await relationalGetSchema.invoke({
  vendor: 'postgresql',
  connectionString: DB_URL,
  tables: ['users', 'orders'],
  cacheTtlMs: 300000,
});

// Success response:
// {
//   success: true,
//   schema: {
//     vendor: 'postgresql',
//     tables: [
//       {
//         name: 'users',
//         schema: 'public',
//         columns: [
//           { name: 'id', type: 'integer', isNullable: false, isPrimaryKey: true, defaultValue: null },
//           { name: 'email', type: 'varchar(255)', isNullable: false, isPrimaryKey: false, defaultValue: null },
//         ],
//         primaryKey: ['id'],
//         foreignKeys: [],
//         indexes: [{ name: 'users_pkey', columns: ['id'], isUnique: true }],
//       },
//     ],
//     generatedAt: '2026-02-21T...',
//   },
//   summary: { tableCount: 2, columnCount: 8, foreignKeyCount: 1, indexCount: 3 },
// }

Parameter	Type	Required	Description
vendor	string	Yes	Database vendor
connectionString	string	Yes	Connection string
database	string	No	Database name for cache scoping
tables	string[]	No	Filter to specific tables
cacheTtlMs	number	No	Cache TTL in ms (0 disables)
refreshCache	boolean	No	Force cache invalidation

ConnectionManager

The ConnectionManager class handles connection lifecycle, pooling, and reconnection.

import { ConnectionManager } from '@agentforge/tools';

const manager = new ConnectionManager(
  {
    vendor: 'postgresql',
    connection: {
      connectionString: 'postgresql://user:pass@localhost:5432/mydb',
      pool: { max: 20, acquireTimeoutMillis: 5000, idleTimeoutMillis: 30000 },
    },
  },
  { enabled: true, maxAttempts: 5, baseDelayMs: 1000, maxDelayMs: 30000 },
);

Methods:

Method	Returns	Description
connect()	Promise<void>	Connect to the database (idempotent)
disconnect()	Promise<void>	Close connection and cancel reconnection
dispose()	Promise<void>	Disconnect and remove all event listeners
isConnected()	boolean	Check connection status
getState()	ConnectionState	Get current state ('disconnected', 'connecting', 'connected', 'reconnecting', 'error')
getVendor()	DatabaseVendor	Get configured vendor
getPoolMetrics()	PoolMetrics	Pool stats: totalCount, activeCount, idleCount, waitingCount
isHealthy()	Promise<boolean>	Health check (SELECT 1)

Events:

Event	Payload	Description
'connected'	—	Connection established
'disconnected'	—	Connection closed
'error'	Error	Connection error
'reconnecting'	{ attempt, maxAttempts, delayMs }	Reconnection attempt scheduled

withTransaction

ACID transactions with isolation levels and timeout:

import { withTransaction, ConnectionManager } from '@agentforge/tools';
import { sql } from 'drizzle-orm';

const result = await withTransaction(manager, async (tx) => {
  await tx.execute(sql`UPDATE accounts SET balance = balance - 100 WHERE id = 1`);
  await tx.execute(sql`UPDATE accounts SET balance = balance + 100 WHERE id = 2`);
  return { transferred: true };
}, {
  isolationLevel: 'serializable',
  timeoutMs: 5000,
});

drizzle-orm Dependency

The sql tagged template comes from drizzle-orm. Add it as a direct dependency:

pnpm add drizzle-orm

Option	Type	Description
isolationLevel	'read uncommitted' | 'read committed' | 'repeatable read' | 'serializable'	Transaction isolation level
timeoutMs	number	Max transaction duration before auto-rollback

The TransactionContext (tx) provides: execute(), isActive(), commit(), rollback(), createSavepoint(), rollbackToSavepoint(), releaseSavepoint(), withSavepoint().

File Tools (18)

File Operations

import { 
  fileReader, 
  fileWriter, 
  fileAppend,
  fileDelete,
  fileExists 
} from '@agentforge/tools';

// Read file
const readResult = await fileReader.invoke({
  path: '/path/to/file.txt',
  encoding: 'utf-8'
});
if (readResult.success) {
  console.log(readResult.data?.content); // File contents
  console.log(readResult.data?.size); // File size in bytes
  console.log(readResult.data?.path); // '/path/to/file.txt'
  console.log(readResult.data?.encoding); // 'utf-8'
}

// Write file
await fileWriter.invoke({
  path: '/path/to/file.txt',
  content: 'Hello World'
});

// Append to file
await fileAppend.invoke({
  path: '/path/to/file.txt',
  content: '\nNew line'
});

// Delete file
await fileDelete.invoke({
  path: '/path/to/file.txt'
});

// Check if file exists
const exists = await fileExists.invoke({
  path: '/path/to/file.txt'
});

Directory Operations

import { 
  directoryList, 
  directoryCreate,
  directoryDelete,
  fileSearch 
} from '@agentforge/tools';

// List directory
const listResult = await directoryList.invoke({
  path: '/path/to/dir',
  recursive: true
});
if (listResult.success) {
  console.log(listResult.data?.files); // Array of file objects
  console.log(listResult.data?.count); // Number of files
  console.log(listResult.data?.path); // '/path/to/dir'
}

// Create directory
await directoryCreate.invoke({
  path: '/path/to/new/dir'
});

// Search files
const searchResult = await fileSearch.invoke({
  directory: '/path/to/dir',
  pattern: '*.ts',
  recursive: true
});
if (searchResult.success) {
  console.log(searchResult.data?.matches); // Array of matching file paths
  console.log(searchResult.data?.count); // Number of matches
  console.log(searchResult.data?.directory); // '/path/to/dir'
  console.log(searchResult.data?.pattern); // '*.ts'
}

Filesystem Confinement

The standalone file and directory tool exports are privileged filesystem operations. For model-exposed agents, configure both factory groups with one shared policy:

import {
  createDirectoryOperationTools,
  createFileOperationTools,
  createFileSystemPolicy,
} from '@agentforge/tools';

const policy = createFileSystemPolicy({
  workspaceRoot: process.cwd(),
});

const safeFileTools = createFileOperationTools({ policy });
const safeDirectoryTools = createDirectoryOperationTools({ policy });

workspaceRoot resolves relative paths from the workspace and confines all configured operations to that root. Use allowedRoots for multiple permitted roots. The policy rejects lexical traversal, resolved symlink escapes, paths outside the allowed roots, and recursive deletion of an allowed root. Set allowOutsideRoots: true only for trusted local automation that explicitly needs unrestricted host access.

Model-Safe File and Web Preset

When file and web tools are exposed to a model, use the combined preset instead of wiring separate policy objects by hand:

import { createModelSafeToolPreset } from '@agentforge/tools';

const preset = createModelSafeToolPreset({
  fileSystem: { workspaceRoot: process.cwd() },
});

const agentTools = preset.tools;

The preset requires workspaceRoot or allowedRoots, applies one confinement policy to all file and directory tools, and preserves the default web blocks for localhost, private-network, link-local, metadata, and unsafe redirect destinations. Privileged override flags are forced back to safe values; use the standalone factories only for trusted operator-controlled automation.

Utility Tools (22)

Date/Time

import { dateFormatter, currentDateTime, dateArithmetic, dateDifference } from '@agentforge/tools';

String Utilities

import {
  stringCaseConverter,
  stringTrim,
  stringReplace,
  stringSplit,
  stringJoin,
  stringSubstring,
  stringLength
} from '@agentforge/tools';

Math Operations

import {
  calculator,
  mathFunctions,  // Includes round, abs, floor, ceil, etc.
  randomNumber,
  statistics
} from '@agentforge/tools';

Validation

import {
  emailValidator,
  urlValidatorSimple,
  phoneValidator,
  creditCardValidator,
  ipValidator,
  uuidValidator
} from '@agentforge/tools';

Agent Tools (1)

Ask Human

import { createAskHumanTool } from '@agentforge/tools';

// Create the tool
const askHuman = createAskHumanTool();

// Use in your agent
const tools = [askHuman, ...otherTools];

// The agent can call this tool to request human input
// Execution pauses until a human responds

Features:

• Human-in-the-loop workflows
• Priority levels (low, normal, high, critical)
• Timeout handling with default responses
• Suggested responses for UI
• LangGraph interrupt integration

Note: Requires @langchain/langgraph to be installed.

Complete Tool List

See the Tools README for the complete list of all 88 tools with detailed documentation.