Tools

OpenCode

Open-source terminal coding agent, running on Omnizen models.

1. Get an API key

Go to your dashboard → API keys, click Create API key, and copy the sk-… token. You only see it once — save it somewhere safe.

2. Install OpenCode

Easiest path on macOS, Linux (glibc), or Windows:

bash

npm install -g opencode-ai

See the official OpenCode docs for native installers and shell completions.

Heads-up for Alpine / musl users: OpenCode ships a glibc binary. Use Ubuntu, Debian, or any glibc-based image instead.

3. Configure Omnizen as a provider

Create or edit ~/.config/opencode/config.json (Windows: %APPDATA%\opencode\config.json):

~/.config/opencode/config.jsonjson

{
  "$schema": "https://opencode.ai/config.json",
  "provider": {
    "omnizen": {
      "npm":  "@ai-sdk/openai-compatible",
      "name": "Omnizen",
      "options": {
        "baseURL": "https://api.omnizen.ai/v1",
        "apiKey":  "sk-…paste-your-omnizen-key-here"
      },
      "models": {
        "deepseek-flash":              { "name": "DeepSeek Flash" },
        "deepseek-pro":                { "name": "DeepSeek Pro (reasoning)" },
        "kimi-k2.6":                   { "name": "Kimi K2.6 (262K context)" },
        "claude-3-5-haiku-20241022":   { "name": "Claude 3.5 Haiku alias" },
        "claude-3-5-sonnet-20241022":  { "name": "Claude 3.5 Sonnet alias" }
      }
    }
  }
}

Add or remove models as you like — anything in your Omnizen catalog works. The provider is OpenAI-compatible, so OpenCode talks to https://api.omnizen.ai/v1/chat/completions with your key as a Bearer token.

4. Smoke-test it

bash

opencode run --model omnizen/deepseek-flash "what is 2+2?"

You should get back something like:

text

> build · deepseek-flash

4

Try Kimi too:

bash

opencode run --model omnizen/kimi-k2.6 "what is 7*8? reply with only the number"

5. Open the full TUI

bash

opencode

Inside, hit /models to switch between any models you defined in the config. Tokens stream live as the model generates them — every call is also logged on your dashboard.

Tips

Reasoning models need bigger budgets. Models like kimi-k2.6 internally consume tokens to "think" before replying — set --max-tokens high enough (e.g. 1500+) or the visible reply may be empty.
Per-project config. Drop a project-local opencode.json at your repo root to scope a single codebase to a dedicated Omnizen key. The file format is identical.
Streaming + tool calling work. No special flags — OpenCode's agent flows (file edits, shell exec, multi-step plans) all run through Omnizen the same way they would against OpenAI.
Same key, multiple tools. The exact same sk-… token works for Cline, Codex CLI, Aider, Continue, Cursor, and any OpenAI-compatible client.