Epho

What is Epho?

Epho is a cloud-based API service that runs AI coding agents like Claude Code, Codex, or OpenCode on your behalf. Instead of installing and managing these agents locally, you send a simple HTTP POST request with a prompt, and Epho handles everything else — booting a sandbox, cloning your repositories, connecting your tools, and streaming the agent's work back to you in real time. There's no SDK to integrate, no daemon to run, and no infrastructure to maintain on your side.

Who it's for

• Backend developers who want to automate coding tasks without managing local agent environments or juggling multiple CLI tools.
• Teams building AI-powered workflows that need to trigger agent runs programmatically — from CI pipelines to internal tooling — with a single JSON API call.
• Developers who value flexibility and want to switch between Claude Code, Codex, or OpenCode by changing one field, without rewriting their integration.

Key features

One endpoint, full control

The entire API surface is a single POST /api/v1/chat endpoint. You write JSON with your prompt, choose a harness and model, pass your provider key, and optionally attach repos, files, or MCP servers. Everything else — sandbox creation, cloning, streaming — is handled automatically.

Durable multi-turn conversations

A chat is a persistent session with a sandbox attached. When a turn ends, the box stops; when you send the next prompt with the same chat_id, it resumes with the same filesystem, checkout, and agent session. If the sandbox is gone, a replacement boots and restores the agent's snapshot — the conversation outlives the machine.

Rich context in every request

You can attach up to 32 repositories (with branch and token support), up to 20 input files (via URL or base64), MCP servers, environment variables, a custom system prompt, and an effort level. All of this context rides along with the prompt, so the agent starts with everything it needs.

Bring your own provider key

Your provider API key travels with each request. You pay your provider directly, and Epho never sits between you and the model. For quick experiments, the OpenCode models ending in -free run without any key at all.

What stands out

"You talk HTTP. The agent talks to your repo."

That's the entire product philosophy in one line. Epho collapses the complexity of running AI coding agents — sandboxes, session management, context wiring, event streaming — into a single HTTP call. The design is deliberately minimal: one endpoint, good defaults, and everything else falls out of that. It's the kind of tool that feels obvious once you see it, but only because someone actually built it.

Worth checking out if…

You're building automation around AI coding agents and want to skip the plumbing. If you've ever wrestled with installing Claude Code locally, managing sandboxes, or wiring MCP servers by hand, Epho removes that entire layer. It's especially useful if you need agent runs to be scriptable, resumable, and connected to real repositories — without committing to a specific vendor's agent. And if you value clean APIs and minimal moving parts, this is about as close to "just POST a prompt" as it gets.