Summary

Dexto is an orchestration layer for AI agents, enabling developers to build agentic applications that translate natural language into real-world actions. It connects LLMs, tools, and data into stateful, persistent systems, empowering users to create sophisticated AI assistants, copilots, and multi-agent collaborations with ease.

How It Works

Dexto functions as a universal agent interface, orchestrating a wide array of Large Language Models (LLMs) and external tools via the Model Context Protocol (MCP). Its core design emphasizes statefulness, featuring persistent sessions with saved context and memory, alongside native multimodal capabilities for handling text, images, and files within a single conversation. Agents are customizable via YAML or TypeScript, and the system supports pluggable storage backends like Redis and PostgreSQL for robust persistence.

Quick Start & Requirements

Installation is straightforward via NPM: npm install -g dexto. Alternatively, it can be built from source. Key requirements include Node.js, pnpm, and API keys for desired LLM providers (e.g., OpenAI, Anthropic, Google). A quick start involves running dexto for the interactive CLI or dexto --mode web to launch the Web UI. Official documentation for configuration, building, and API reference is available.

Highlighted Details

• Extensive LLM Support: Integrates over 20 LLMs from major providers (OpenAI, Anthropic, Google, Groq, xAI, Cohere) and local models, allowing dynamic switching.
• Native Multimodal: Seamlessly handles text, images, files, and tool interactions within a unified conversational flow.
• Persistent Sessions & Memory: Conversations, context, and agent memory are saved, enabling continuity and export/import functionality.
• Agent Recipes: Offers pre-built, installable agents for common tasks like image generation, podcast creation, and database operations, simplifying setup for specific use cases.
• Model Context Protocol (MCP): Facilitates integration with over 100 tools and external services, extending agent capabilities.
• Flexible Deployment: Supports local, cloud, and hybrid execution environments, alongside various interfaces including CLI, WebUI, APIs, and bots.

Maintenance & Community

Dexto is developed by the Truffle AI team and welcomes community contributions. A Discord server is available for support, project sharing, and discussion. Anonymous usage data is collected to improve the project, with an opt-out option provided.

Licensing & Compatibility

The project is licensed under the Elastic License 2.0. This license is source-available but imposes restrictions, notably prohibiting its use as a hosted service without specific authorization, which may impact commercial adoption for SaaS offerings.

Limitations & Caveats

The Elastic License 2.0's restrictions on offering Dexto as a hosted service are a significant consideration for commercial deployment. By default, the project collects anonymous telemetry data, which users can opt out of. Setting up integrations with certain LLM providers requires obtaining and configuring API keys.