Dexto
A coding agent and general agent harness for building and orchestrating agentic applications.
Last verified:
What is Dexto?
Dexto is an all-in-one toolkit and universal intelligence layer for building collaborative, context-aware AI Agents and agentic applications that turn natural language into real-world actions. Developed by Truffle AI and backed by Y Combinator, it orchestrates LLMs, tools, and data into persistent, stateful systems with memory, enabling developers to rapidly create AI assistants, digital companions, and copilots that think, act, and feel alive.
Key features include a configuration-driven framework where agent behavior is defined in YAML files (no complex programming required), support for 50+ LLMs including OpenAI, Anthropic, Google, Groq, and local models via Ollama, and native MCP (Model Context Protocol) integration connecting to 100+ tools and services. Dexto provides multiple interfaces including CLI, Web UI, REST API, WebSocket API, and TypeScript SDK, along with persistent sessions that maintain context across conversations, production-ready storage options (Redis, PostgreSQL, SQLite), and human-in-the-loop approval policies for tool execution.
Dexto is designed for developers building AI-native apps, teams needing autonomous coding and research agents, enterprises requiring local-first deployment for privacy and security, and anyone wanting to create agentic applications for coding, research, task execution, document analysis, image editing, podcast generation, music creation, and customer support routing. Pre-built agent recipes include coding-agent, nano-banana-agent (image generation), podcast-agent, github-agent, database-agent, triage-agent, talk2pdf-agent, and more.
Dexto pricing
Pricing model: Freemium
Free and open source under Elastic 2.0 license. No paid tiers mentioned on the website. Users pay only for their own LLM API keys (OpenAI, Anthropic, Google, Groq, etc.) when using cloud models. Local models via Ollama are free. The platform itself is free to install via npm (npm install -g dexto) or build from source. Nova credits are available for purchase as an alternative to bringing your own API keys.
Dexto pros
- Configuration-driven YAML framework - define agent behavior without complex programming
- Support for 50+ LLMs including OpenAI, Anthropic, Google, Groq, and local models
- MCP integration connects to 100+ tools via Model Context Protocol out of the box
- Local-first design - run entirely on your infrastructure with no cloud required
- Persistent sessions maintain conversation context and memory across interactions
- Multiple interfaces: CLI, Web UI, REST API, WebSocket API, TypeScript SDK
- Pre-built agent recipes for coding, podcasts, images, GitHub, database, and more
- Human-in-the-loop approval policies for safe tool execution
- Production-ready storage with Redis, PostgreSQL, SQLite support
- Built-in OpenTelemetry distributed tracing and observability
- From idea to running agent in under 5 minutes with no setup required
- Full debugging and real-time agent monitoring capabilities
- Swap models and tools instantly by editing YAML config without touching code
- Open source under Elastic 2.0 license
- Y Combinator backed with active developer community on Discord
- Native multimodal support for text, images, files, and tools in single conversation
- Pluggable storage backends for cache, database, and blob storage
- Can expose agents as MCP servers for use in Cursor and other tools
- Docker support for simple deployments and portability
Dexto cons
- Currently in Beta testing phase with potential instability
- Requires API keys for most LLM providers (costs not included)
- Beta version means features may change over time
- Learning curve for YAML configuration despite being declarative
- MCP server setup required for many external tools
- Self-hosting requires infrastructure management responsibility
- Analytics collection enabled by default (must opt-out manually)
- Limited documentation compared to mature frameworks
- 329 GitHub stars indicates relatively small community
- Elastic 2.0 license restricts certain commercial use cases
- Some advanced features require TypeScript knowledge
Frequently asked questions about Dexto
What is Dexto?
Dexto is a universal agent intelligence layer for building collaborative, context-aware AI Agents and agentic apps. It orchestrates LLMs, tools, and data into persistent, stateful systems with memory, so you can rapidly create AI assistants, digital companions, and copilots that think, act, and feel alive. It combines a configuration-driven framework, robust runtime, and seamless developer experience.
How do I get started with Dexto?
Install the CLI with npm install -g dexto or curl -fsSL https://dexto.ai/install | bash. Run 'dexto' to start the setup workflow which prompts for your preferred LLM and API keys and opens the Web UI. Alternatively start interactive CLI mode with 'dexto --mode cli'.
Is Dexto free to use?
Yes, Dexto is open source under the Elastic 2.0 license and free to install. You only pay for your own LLM API keys when using cloud models like OpenAI, Anthropic, or Google. Local models via Ollama are completely free.
Can I run Dexto locally without cloud dependency?
Yes, Dexto is fully local with no cloud required. You can self-host on your own infrastructure for complete control, privacy, and offline operation. This local-first design ensures you own your data and makes Dexto ideal for enterprise environments with strict security needs.
What LLMs does Dexto support?
Dexto supports 50+ LLMs including OpenAI (gpt-5, gpt-4o, o3, o1), Anthropic (claude-sonnet-4-5, claude-opus-4-1), Google (gemini-2.5-pro, gemini-2.5-flash), Groq (llama-3.3-70b, qwen3-32b), xAI (grok-4, grok-3), Cohere, and local models via Ollama. You can instantly swap models without touching code.
What agents come pre-built with Dexto?
Dexto includes coding-agent (code generation/refactoring), nano-banana-agent (image generation/editing), podcast-agent (multi-speaker audio), sora-video-agent (AI video), database-agent (SQL queries), github-agent (PR/repo management), image-editor-agent, music-agent, talk2pdf-agent, product-researcher, and triage-agent (customer support routing).
How does MCP integration work in Dexto?
Dexto uses the Model Context Protocol (MCP) standard, connecting to any MCP-compatible tool out of the box including databases, APIs, file systems, and other AI frameworks. You can add custom MCP servers by editing your agent YAML or using the WebUI. The orchestration layer handles all integration complexity.
What interfaces are available for interacting with Dexto?
Dexto provides Web UI (default chat interface with image support), Interactive CLI for terminal automation, Headless Server mode for REST and WebSocket APIs, MCP Server mode for exposing agents as tools, Discord Bot mode, Telegram Bot mode, and a TypeScript SDK for programmatic integration.
How does session management work in Dexto?
Dexto maintains persistent sessions with conversation memory saved to storage backends (Redis, PostgreSQL, SQLite, or in-memory). You can create, list, delete, and search sessions. Sessions maintain context across conversations and can be exported, imported, or shared across environments.
What is the human-in-the-loop feature?
Human-in-the-loop allows you to configure approval policies for tool execution. Agents can remember which tools are safe per session and ask for confirmation before executing potentially risky actions. Use --auto-approve flag to bypass confirmations when fast local iteration is needed.