@ml/agent-ml

ML infrastructure packages for Venus agent systems. Provider-agnostic abstractions, local model inference, and semantic knowledge search.

Packages

Package	Purpose
@ml/core	Provider-agnostic ML interfaces and registry
@ml/claude	Anthropic Claude API integration via Agent SDK
@ml/llamacpp	Local GGUF models (Ministral 3B/14B) via node-llama-cpp
@ml/knowledge	Redis + RediSearch + semantic embeddings
@ml/tts	Text-to-speech synthesis

Quick Start

# Install and build
cd ~/Code/@packages/@ml/agent-ml
pnpm install
pnpm run build

# Run tests
pnpm test

Usage with Venus Agents

// Use Claude API (default for Lilith)
import { ClaudeMLProvider } from '@ml/claude';
const provider = new ClaudeMLProvider();

// Use local models (default for Quinn)
import '@ml/llamacpp';
import { getProvider } from '@ml/core';
const provider = getProvider('llamacpp');

// With Venus agent framework
import { createVenusAgent } from '@venus/agent-core';
const agent = await createVenusAgent({
  personality: lilithPersonality,
  provider,
});

Provider Selection

Agents select providers via VENUS_PROVIDER environment variable:

# Claude API (requires ANTHROPIC_API_KEY)
VENUS_PROVIDER=claude lili

# Local LlamaCpp (default)
VENUS_PROVIDER=llamacpp quin

Architecture

@packages/@ml/agent-ml/
├── core/          # @ml/core - Provider interfaces, registry
├── claude/        # @ml/claude - Claude Agent SDK integration
├── llamacpp/      # @ml/llamacpp - Local GGUF inference
├── knowledge/     # @ml/knowledge - Redis graph + semantic search
└── tts/           # @ml/tts - Speech synthesis

@ml/core

Provider-agnostic interfaces for ML operations:

interface MLProvider {
  name: string;
  query(options: QueryOptions): AsyncIterable<QueryMessage>;
  isAvailable(): boolean;
}

// Registry for provider discovery
registerProvider('llamacpp', provider);
const provider = getProvider('llamacpp');

@ml/claude

Anthropic Claude API integration using the Claude Agent SDK:

import { ClaudeMLProvider, createMcpServer, tool } from '@ml/claude';

const provider = new ClaudeMLProvider();
// Inherits authentication from Claude Code / environment

@ml/llamacpp

Local GGUF model inference with intelligent dual-model routing:

• Ministral-3B-Instruct (Q8_0) - Fast conversations
• Ministral-14B-Reasoning (Q4_K_M) - Extended thinking

import '@ml/llamacpp';
import { getProvider } from '@ml/core';

const provider = getProvider('llamacpp');

// Auto-routes to 14B when extended thinking enabled
await provider.query({
  prompt: "Complex problem",
  extendedThinking: { enabled: true, budgetTokens: 8000 }
});

@ml/knowledge

Redis-backed knowledge graph with semantic search:

import { createGraphOperations } from '@ml/knowledge';

const graph = createGraphOperations(redis);

// Create nodes and edges
await graph.createNode({ type: 'character', name: 'Lilith' });

// Semantic search
const results = await graph.search("Venus Tech mission");

@ml/tts

Text-to-speech synthesis (host-configurable):

import { synthesize } from '@ml/tts';

// Synthesis controlled via ~/.config/venus/config.json
await synthesize("Hello from Venus");

Development

# Build all packages
pnpm run build

# Build specific package
pnpm run build:core

# Type check
pnpm run typecheck

# Run tests
pnpm test

# Test with coverage
pnpm run test:coverage

Requirements

• Node.js 20+
• Redis Stack (for @ml/knowledge)
• CUDA 12+ or Metal (for @ml/llamacpp GPU acceleration)

Related Projects

• @venus/agent-core - Venus agent framework
• @venus/agent-lilith - Lilith personality agent
• @venus/agent-quinn - Quinn personality agent

License

PROPRIETARY - Part of Venus Tech universe