Epho

What is Epho?

Epho turns coding agents into an API. Send one POST request with a prompt, and it boots a cloud sandbox running Claude Code, Codex, or OpenCode with your repos cloned and MCP tools connected. The agent's work streams back live, and chats can resume later in the same environment with full context.

Top Features:

• Single endpoint: one JSON body sets the harness, model, prompt, repos, and files.
• Resumable chats: pass a chat ID to continue in the same sandbox.
• Async runs: stream events over SSE or collect results later by webhook.

Use Cases:

• PR reviews: trigger an agent from CI to review every new pull request.
• Bug fixing: describe a failing test and let the agent patch the repo.
• Scheduled jobs: run agent tasks from cron or bash scripts over plain HTTP.

Who Can Use Epho?

• Backend developers: automate coding tasks without installing or running agents on local machines.
• Platform teams: wire coding agents into CI pipelines and internal tools over HTTP.
• Product builders: add agent features, like app generators or reviewers, to products.

Pricing

• Starting credit (free): ten dollars on signup, roughly 60 hours of the default instance.
• Usage (about $0.0026 per minute): compute, memory, and disk are billed only during active runs.
• Model tokens (your provider price): tokens bill directly to your own Anthropic, OpenAI, or OpenCode account.

Pros and Cons

Pros:

• Tiny integration: setup is one API key and a single curl request.
• No token markup: your provider key rides along, so model costs stay unchanged.
• Fault tolerance: runs re-queue on a fallback sandbox if a machine dies.

Cons:

• Developer only: no visual interface, so non-coders will struggle to use it.
• Separate bills: you pay compute here and model tokens to your provider.
• Early product: launched in 2026, so integrations and community resources are limited.

FAQs:

1) Which coding agents can it run?

Claude Code, OpenAI Codex, and open-source OpenCode, switchable with one field.

2) Do I need my own model API key?

Usually yes, though OpenCode models ending in free need no key.

3) How much does a typical run cost?

A one-minute run on the default instance costs about a quarter cent.

4) Can I attach private repositories?

Yes, up to 32 repos per request, each with its own token.

5) Can the agent use my tools?

Yes, connect local or remote MCP servers such as Linear or Figma.