Coding agents on ozeye

Every popular coding agent can run on EU-hosted open models through ozeye. Two protocols cover all of them: the OpenAI-compatible API at https://ozeye.ai/v1, and the Anthropic Messages API at https://ozeye.ai for Claude Desktop, Cowork and Claude Code. One key from Dashboard → API keys works for both.

Anthropic protocol

Claude Code and Claude Desktop & Cowork speak their native protocol to ozeye. No proxy, no shim.

OpenAI protocol

Cursor, Hermes, OpenClaw and OpenCode use a custom OpenAI-compatible provider. Codex works only up to version 0.94.0.

Which model

glm-5.3opus tierBest tool use and long agentic sessions; the default we recommend for agents.
deepseek-v4.1-flashsonnet tierEveryday coding at flash speed, native vision; 1M context.
deepseek-v4-flash-0731haiku tierFast turns, subagents, titles and summaries; hosted by three providers.
kimi-k2.6opus tier, optionalAlternative selectable by id; 262k context.

Model ids are exactly the ids on the models page. There are no vendor prefixes and no aliases to remember.

What you get on every agent

  • EU hosting and zero data retention: prompts and source code are not stored by ozeye, see data retention.
  • One balance and per-key spend limits, so a runaway agent cannot exceed what you allowed.
  • Automatic failover between providers of the same model where more than one hosts it.
  • Every request in Dashboard → Logs within seconds, with cost per request.