An open-source AI coding agent that runs in your terminal, with companion desktop, web, editor, and chat interfaces.
truffle-ai/dexto
An open-source agent harness in TypeScript for building stateful LLM agents, with a built-in coding agent, CLI, Web UI, and SDKs.
- Stars
- 657
- Forks
- 87
- Open issues
- 78
- Last commit
- 6 days ago
What is it?
Dexto is an agent harness: an orchestration layer that handles session state, tool calls, memory, and error recovery so that an LLM can act as a reliable agent. Agents are defined in YAML, so models, MCP servers, storage, and permissions can be changed without modifying code.
The project ships a coding agent that can be used from the CLI or a Web UI to build apps, read and refactor code, run shell commands, and spawn sub-agents for exploration. It supports 50+ LLMs across OpenAI, Anthropic, Google, local models via Ollama or GGUF, and gateways such as OpenRouter and AWS Bedrock, with model switching mid-conversation.
Dexto can run as a CLI, a Web UI, a REST and SSE server, an MCP server over stdio or HTTP, or as an embedded library through the @dexto/core TypeScript SDK. Discord and Telegram integrations are provided as examples.
Key features
- 01Defines agents in version-controlled YAML, with hot reload of config changes.
- 02Connects to MCP servers and exposes agents as MCP servers.
- 03Switches between 50+ LLMs, including local models, during a conversation.
- 04Persists sessions and memory, with conversation search.
- 05Requires tool approvals in manual mode or auto-approves for trusted use.
- 06Supports Redis, PostgreSQL, and SQLite storage backends.
Use cases
- Running a terminal or web-based coding agent that edits and tests a codebase.
- Building a domain-specific agent by writing a YAML config and connecting MCP tools.
- Exposing an agent backend through a REST and SSE API for a web frontend.
- Deploying a chat bot on Discord or Telegram backed by a configurable agent.
Quick start
- 1
Run the official install script from the README (macOS, Linux, WSL)
- 2
Or clone the repository to build from source
git clone https://github.com/truffle-ai/dexto.git - 3
Install dependencies and the CLI from the repo folder
cd dexto && pnpm install && pnpm install-cli - 4
Start Dexto; the setup wizard runs on first launch
dexto
Things to know
- The README marks the project as Beta, so breaking changes are possible.
- Cloud models need your own provider API key, such as OpenAI or Anthropic.
- The license is shown as Elastic 2.0 in the README badge, but the project metadata does not clearly specify a license.
Topics
FAQ
- What is dexto?
- An open-source agent harness in TypeScript for building stateful LLM agents, with a built-in coding agent, CLI, Web UI, and SDKs. Dexto is an agent harness: an orchestration layer that handles session state, tool calls, memory, and error recovery so that an LLM can act as a reliable agent. Agents are defined in YAML, so models, MCP servers, storage, and permissions can be changed without modifying code.
- Is dexto free to use? What license does it have?
- dexto is open source under the Other license. Check the license terms before using it in commercial products.
- Is dexto still maintained?
- Actively maintained. The last commit to dexto was on Oct 16, 2026.
- What can I use dexto for?
- Running a terminal or web-based coding agent that edits and tests a codebase. Building a domain-specific agent by writing a YAML config and connecting MCP tools. Exposing an agent backend through a REST and SSE API for a web frontend. Deploying a chat bot on Discord or Telegram backed by a configurable agent.
- How do I install dexto?
- Run the official install script from the README (macOS, Linux, WSL) Or clone the repository to build from source: git clone https://github.com/truffle-ai/dexto.git Install dependencies and the CLI from the repo folder: cd dexto && pnpm install && pnpm install-cli Start Dexto; the setup wizard runs on first launch: dexto
- Where can I get dexto?
- The source code and installation instructions are on GitHub: https://github.com/truffle-ai/dexto
Summary generated by AI from the project's README. GitHub stats refresh daily.
Related projects
An open-source coding agent that reads your project, edits files and runs tests from your terminal, desktop, browser or editor.
A CLI-first open-source AI coding agent runtime that works with any model via MCP tools, interactively, piped, as a daemon, or over WebSocket/ACP.
Open-source MCP host desktop app that connects any function-calling LLM, such as ChatGPT, Claude, or Ollama models, to MCP tools.
Self-hosted personal AI agent framework in Python with a browser WebUI, terminal client, chat app channels, tools, memory and MCP support.
Open-source SDK for building and running AI agents in Python and TypeScript, with tools, MCP, memory, and model portability.