Happy slice of mcp pizzaMCP.pizza

saiki

MCP.Pizza Chef: truffle-ai

Formerly called Saiki, Dexto is an open harness for running agents on your own machine. It ships with a ready-made coding agent and lets you define others in a plain settings file. You can switch between more than fifty models mid-conversation, including local ones, connect any MCP server for extra tools, and decide which actions need your approval before they happen. Conversations survive a restart. Use it in a terminal, a browser chat window, Discord, or Telegram. Bring your own model key. Still labelled beta.

Coding

Use This MCP client To

Have an agent build a small app from a one-line description Switch models mid-conversation without losing what we discussed Approve each risky action before the agent takes it Pick up a coding session from yesterday where I left it Run an agent on a local model so nothing leaves my machine Add tools from an MCP server to an agent I defined myself

README

Dexto

Deutsch | English | Español | français | 日本語 | 한국어 | Português | Русский | 中文

An open agent harness for AI applications—ships with a powerful coding agent.

Dexto Demo

What is Dexto?

Dexto is an agent harness—the orchestration layer that turns LLMs into reliable, stateful agents that can take actions, remember context, and recover from errors.

Think of it like an operating system for AI agents:

Component Analogy Role
LLM CPU Raw processing power
Context Window RAM Working memory
Dexto Operating System Orchestration, state, tools, recovery
Your Agent Application Domain-specific logic and clients

Why Dexto?

  • Configuration-driven: Define agents in YAML. Swap models and tools without touching code.
  • Batteries included: Session management, tool orchestration, memory, multimodal support, and observability—out of the box.
  • Run anywhere: Local, cloud, or hybrid. CLI, Web UI, REST API, Discord, Telegram, or embedded in your app.

What You Can Build

saiki FAQ

Do I need a key?
For hosted models, yes. A setup wizard walks you through it, and you can run local models instead if you would rather keep everything on your own machine.
Can I use this to build a whole app?
Yes. The bundled coding agent can create files, run commands, execute tests, and hand parts of the work to helper agents.
Where do I use it?
In a terminal, in a browser chat window, or connected to Discord or Telegram. It can also be embedded in an app you are building.
Which models does it support?
More than fifty, across OpenAI, Anthropic, Google, Groq, and xAI, plus local models through Ollama and gateways such as OpenRouter and Bedrock.
Is it finished?
No. The makers label it beta, so expect rough edges and changes between versions.
How hard is setup?
You run one install command in a terminal and then a wizard. Defining your own agents means editing a plain text settings file.
Can it use MCP servers?
Yes. You list the servers you want in an agent's settings file, or browse and add them from the built-in store in the web interface.