spec-ai-core

Core functionality for the spec-ai framework.

Overview

This crate provides the foundational components for building AI agents, including:

• Agent Runtime: Core agent execution engine and lifecycle management
• Tool System: Extensible tool framework for agent capabilities
• Embeddings: Vector embeddings for semantic search and similarity
• Provider Integrations: Support for multiple LLM providers
• CLI Helpers: Terminal UI components and utilities

Features

The crate supports multiple LLM providers and capabilities through feature flags:

LLM Providers

• openai - OpenAI API integration
• anthropic - Anthropic Claude API integration
• ollama - Ollama local model support
• mlx - Apple MLX framework integration
• lmstudio - LM Studio local model support

Additional Features

• vttrs - Video/subtitle processing support
• web-scraping - Web scraping capabilities via Spider
• api - HTTP API functionality
• integration-tests - Integration test support

Dependencies

This crate depends on:

• spec-ai-config - Configuration management
• spec-ai-policy - Policy enforcement

Platform-Specific Behavior

On non-macOS platforms, the extractous dependency is included for document extraction using GraalVM/Tika. This is excluded on macOS due to AWT compatibility issues.

Usage

This is an internal crate primarily used by:

• spec-ai-cli - The command-line interface
• spec-ai-api - The HTTP API server
• spec-ai - The public library crate

For end-user documentation, see the main spec-ai README.