truffle-ai/dexto

What it solves

Dexto provides an orchestration layer (an "agent harness") that transforms raw Large Language Models (LLMs) into reliable, stateful agents. It solves the complexity of managing agent state, tool integration, memory, and error recovery, allowing developers to build autonomous agents without writing extensive boilerplate code.

How it works

Dexto acts as an operating system for AI agents. It uses a configuration-driven approach where agents are defined in YAML files, specifying their LLM provider, tools, and system prompts. It integrates with the Model Context Protocol (MCP) to connect to various external tools and APIs. It supports persistent sessions and memory, enabling agents to remember context across restarts. It also allows for multi-agent systems where a parent agent can spawn specialized sub-agents to handle complex tasks.

Who it’s for

Developers and AI engineers who want to build, deploy, and manage AI agents—ranging from coding assistants to digital companions—across different interfaces (CLI, Web UI, REST API) and LLM providers.

Highlights

• Configuration-driven: Define and swap models and tools via YAML without changing code.
• Extensive LLM Support: Compatible with 50+ models from providers like OpenAI, Anthropic, Google, Groq, and local models via Ollama or node-llama-cpp.
• MCP Integration: Built-in support for Model Context Protocol servers to easily add tools like Puppeteer or filesystem access.
• Production-ready Coding Agent: Ships with a pre-built agent capable of building, refactoring, and debugging code autonomously.
• Human-in-the-Loop: Fine-grained permission controls to require approval for specific tool executions.
• Multi-Agent Orchestration: Ability to spawn ephemeral sub-agents for specialized subtasks.
• Flexible Deployment: Can be run as a Web UI, CLI, or a server exposing REST and SSE APIs.