Epho

Epho packages coding agents as a plain HTTP API. You POST a message to one endpoint, POST /api/v1/chat, and the work streams back over SSE — the agent runs on Epho's machines, not yours, so there is no SDK to install, no long-lived daemon to supervise and no container fleet to keep patched.

What it runs

Three agents are supported today: Claude Code, OpenAI's Codex, and the open-source OpenCode. Each run can clone up to 32 repositories, each with its own auth token, accept up to 20 input files, and mount MCP servers (local or remote) so the agent reaches your own tools. Artifacts come back as presigned download URLs. Sessions are persistent, so a conversation keeps its context across turns and is restored automatically if the underlying infrastructure fails mid-run. You can stream the run live or fire it asynchronously and receive a webhook callback.

Pricing

Billing is per second of compute rather than per seat: roughly $0.0000164 per vCPU-second, $0.0000053 per GiB-second of memory and $0.000000036 per GiB-second of disk. On the default 2 vCPU / 2 GiB / 10 GiB instance a one-minute run works out to about $0.0026. Model tokens are billed separately by whichever provider you bring credentials for; OpenCode models whose names end in -free need no provider key at all. New accounts start with $10 in credit.

Who it is for

Backend teams that want agentic coding inside an existing product or pipeline — a CI job that opens a pull request, a support tool that patches a repo, a SaaS feature that edits customer code — without operating agent sandboxes themselves. Built by Bruin Data Limited.