A Go-based AI Agent foundation library with unified multi-LLM provider interfaces, built-in tool-calling system, and extensible Agent loops for building LLM-driven applications.
Dive is a Go-based AI Agent foundation library developed by Deep Noodle, providing a complete low-level abstraction for building AI Agents and LLM-driven applications. The project requires Go 1.25.0 with the module path github.com/deepnoodle-ai/dive.
Multi-LLM Provider Support: Unified adapters for 7+ providers including Anthropic, OpenAI, Google, Grok, OpenRouter, Mistral, and Ollama, all with tool calling support. Some providers are distributed as independent Go modules for dependency isolation.
Tool-Calling System: 11 built-in tools (Read, Write, Edit, Glob, Grep, ListDirectory, TextEditor, Bash, WebFetch, WebSearch, AskUserQuestion) aligned with Claude Code patterns. Three custom tool approaches: FuncTool (stateless, auto-schema from struct tags), TypedTool[T] (stateful), and Toolset (runtime-dynamic, e.g., MCP servers).
Agent Loop & Extension Mechanism: Automatically runs generate-call-repeat loops with 7 lifecycle hooks (PreGeneration, PostGeneration, PreIteration, PreToolUse, PostToolUse, PostToolUseFailure, Stop) for non-intrusive extension.
Multimodal & Streaming: Supports Text, Image, Document (PDF), ToolUse, ToolResult, Thinking (chain-of-thought), and Refusal content types. Streaming text deltas and tool call events via WithEventCallback.
Session & Interaction: In-memory and JSONL file-based persistent sessions with per-call override; Dialog interface for user interaction abstraction; Skills system for Markdown-based modular capability extension; Suspend & Resume for mid-execution pausing.
Design Principle: The library injects no hidden system prompts or implicit behavior—full user control.
Experimental features (no stability guarantees) include Compaction, Subagent, Sandbox (Docker/Seatbelt), MCP client, CLI (Claude Code-like), and extended toolkits. The companion ecosystem includes Wonton (CLI app framework) and Workflow (multi-step workflow orchestration library).