Skip to content

API.md

Latest commit

 

History

History
286 lines (201 loc) · 7.78 KB

API.md

File metadata and controls

286 lines (201 loc) · 7.78 KB
Raw

FusionPact API Reference

Complete API documentation for FusionPact v2.0.0 Built by FusionPact Technologies Inc.

Table of Contents

Factory Function

create(config?)

Creates a fully-configured FusionPact instance with sensible defaults.

Parameters:

Param Type Default Description
config.embedder string 'mock' 'ollama', 'openai', or 'mock'
config.llmProvider string null 'ollama', 'openai', 'anthropic'
config.enableHybrid boolean true Enable HybridRetriever
config.enableMemory boolean true Enable AgentMemory
config.enableMCP boolean false Start MCP server
config.collection string 'default' Default RAG collection
config.weights object {vector:0.4, tree:0.4, keyword:0.2} Retrieval weights

Returns: { engine, embedder, treeIndex, rag, retriever, memory, orchestrator, mcp }

const fp = require('fusionpact').create({ embedder: 'ollama', enableHybrid: true });

FusionEngine

Core database engine managing collections, CRUD, and multi-tenancy.

new FusionEngine(config?)

Config Type Default Description
dataDir string null Persistence directory
autoSave boolean false Auto-persist

Methods

createCollection(name, config?) → {name, config}

Creates a named vector collection.

Param Type Default
dimensions number 768
distanceMetric string 'cosine'
M number 16
efConstruction number 200
efSearch number 50

insert(collection, entries, options?) → VectorEntry[]

Insert vectors into a collection.

search(collection, queryVector, options?) → SearchResult[]

Search for nearest neighbors.

Option Type Default
topK number 10
filter object null
tenantId string null
includeVectors boolean false

tenant(collection, tenantId) → TenantProxy

Create a tenant-scoped proxy with automatic isolation.

listCollections() → Array

getCollection(name) → object|null

deleteCollection(name) → boolean

get(collection, id) → VectorEntry|null

delete(collection, id) → boolean

exportData() → object

importData(data, options?) → void

HNSWIndex

Low-level HNSW approximate nearest neighbor index.

new HNSWIndex(dimensions, config?)

Methods

Method Returns Description
insert(id, vector, metadata?) VectorEntry Insert a vector
insertBatch(entries) VectorEntry[] Batch insert
search(queryVector, options?) SearchResult[] K-NN search
delete(id) boolean Delete by ID
get(id) VectorEntry|null Get by ID
has(id) boolean Check existence
clear() void Clear all data
serialize() object Export for persistence
HNSWIndex.deserialize(data) HNSWIndex Import from data

TreeIndex

Hierarchical document index for reasoning-based retrieval.

new TreeIndex(config?)

Config Type Default
llmProvider LLMProvider null
maxDepth number 5
maxTokensPerNode number 20000
generateSummaries boolean true

Methods

indexDocument(docId, content, options?) → Promise<TreeNode>

Build a tree structure from a document.

Option Type Default
format string 'text' ('markdown', 'html', 'text')
metadata object {}
title string auto-detected

search(docId, query, options?) → Promise<TreeSearchResult[]>

Reasoning-based retrieval within a document.

searchAll(query, options?) → Promise<TreeSearchResult[]>

Search across all indexed documents.

getTree(docId) → TreeNode|null

listDocuments() → Array

removeDocument(docId) → boolean

HybridRetriever

Multi-strategy retrieval with Reciprocal Rank Fusion.

new HybridRetriever(config)

Config Type Description
engine FusionEngine Required
treeIndex TreeIndex Optional
embedder BaseEmbedder Optional
weights object {vector:0.4, tree:0.4, keyword:0.2}
rrfK number RRF constant (default: 60)

Methods

retrieve(query, options?) → Promise<HybridResult[]>

Option Type Default Description
collection string — Vector collection
docId string — Document for tree search
topK number 10 Max results
strategy string 'hybrid' 'hybrid', 'vector', 'tree', 'keyword'
filter object null Metadata filter
tenantId string null Tenant filter

buildContext(results, options?) → string

recordFeedback(query, strategy, quality) → void

getAdaptiveWeights(query) → object

AgentMemory

Multi-type memory system for AI agents.

Methods

Method Description
remember(agentId, memory) Store episodic memory
learn(agentId, content, metadata?) Add semantic knowledge
registerTool(agentId, tool) Register procedural tool
recall(agentId, query, options?) Cross-memory search
searchAll(agentId, query, options?) Flat ranked search
share(agentId, content, metadata?) Share with other agents
addMessage(agentId, threadId, message) Add conversation message
getConversation(agentId, threadId, options?) Get chat history
forget(agentId, options?) Delete memories (GDPR)
getStats(agentId) Memory statistics

AgentOrchestrator

Multi-agent coordination and communication.

Methods

Method Description
registerAgent(config) Register an agent
unregisterAgent(agentId, options?) Remove an agent
listAgents() List all agents
send(message) Send agent-to-agent message
onMessage(agentId, handler) Register message handler
getMessages(agentId) Get pending messages
delegate(fromAgentId, task, options?) Capability-based delegation
collaborativeRecall(query, options?) Cross-agent retrieval
getStats() Orchestration statistics

Embedding Providers

OllamaEmbedder (Recommended — Free, Local, Private)

new OllamaEmbedder({ model: 'nomic-embed-text', baseUrl: 'http://localhost:11434' })

OpenAIEmbedder

new OpenAIEmbedder({ apiKey: 'sk-...', model: 'text-embedding-3-small' })

MockEmbedder (Testing)

new MockEmbedder({ dimensions: 64 })

Common Methods

Method Description
embed(text) Embed single text → number[]
embedBatch(texts) Embed multiple texts
stats Cache and call statistics
clearCache() Clear embedding cache

LLMProvider

Unified LLM interface for tree reasoning.

new LLMProvider({ provider: 'ollama', model: 'llama3.2' })
new LLMProvider({ provider: 'openai', apiKey: 'sk-...' })
new LLMProvider({ provider: 'anthropic', apiKey: 'sk-ant-...' })

Methods

Method Description
complete(prompt, options?) Generate text completion

Built by FusionPact Technologies Inc.