opencode-kiro

OpenCode plugin for Kiro Gateway with dynamic model discovery, credit usage display, and reasoning effort variants.

What it does

At startup, the plugin:

1. Connects to your Kiro Gateway and fetches available models from GET /v1/models
2. Fetches your credit usage from GET /v1/credits
3. Injects discovered models into the opencode kiro provider — with capabilities derived from the gateway response
4. Adds reasoning effort variants for models that support them (e.g., low/medium/high/xhigh/max)
5. Estimates cost per 1M tokens from the gateway's rate_multiplier
6. Never overwrites models you have already configured explicitly

Requirements

• OpenCode 1.17.7 or newer
• A running Kiro Gateway instance (defaults to http://localhost:8000)
• A Kiro API key (ksk_*) — generate one from kiro-cli settings
• For credit usage and cost display: requires jwadow/kiro-gateway#212 (adds /v1/credits endpoint and ksk_* API key passthrough)

Install

Add to opencode.json:

{
  "$schema": "https://opencode.ai/config.json",
  "plugin": ["opencode-kiro@git+https://github.com/marcelloceschia/opencode-kiro.git"]
}

Configuration

The plugin reads from the provider.kiro block. All fields are optional.

{
  "plugin": ["opencode-kiro@git+https://github.com/marcelloceschia/opencode-kiro.git"],
  "provider": {
    "kiro": {
      "options": {
        "baseURL": "https://kiro.example.com/v1",
        "apiKey": "{env:KIRO_API_KEY}"
      }
    }
  }
}

Option	Default	Description
options.baseURL	http://localhost:8000/v1	Gateway base URL
options.apiKey	—	Kiro API key. Supports {env:VAR} syntax.

Authentication

Set your API key as an environment variable:

export KIRO_API_KEY=ksk_your_key_here

Or inline in the config (not recommended for shared configs):

"apiKey": "ksk_your_key_here"

Model discovery

The plugin discovers all models the gateway exposes and maps capabilities from the response:

• context_window → context limit
• max_output_tokens → output limit
• supported_inputs → attachment support, modalities
• reasoning_efforts → reasoning flag + variants (effort levels in model picker)
• rate_multiplier → estimated cost per 1M tokens

Model names are annotated with usage info:

Claude Opus 4.8 [150/1000 · 2.2x · ≈$0.044/1MT]

Reasoning effort variants

Models with reasoning_efforts get variant entries in the model picker:

claude-opus-4.8 → low, medium, high, xhigh, max
claude-sonnet-4.6 → low, medium, high, max

Overriding a discovered model

Any model you define under provider.kiro.models takes precedence over discovery:

{
  "provider": {
    "kiro": {
      "options": { "apiKey": "{env:KIRO_API_KEY}" },
      "models": {
        "claude-sonnet-4.6": {
          "name": "Sonnet (custom)",
          "limit": { "context": 200000, "output": 32000 }
        }
      }
    }
  }
}

Commands

• /kiro-quota — Check your current Kiro credit usage (plan, credits used/limit, overage, reset date)

Error handling

All network calls degrade gracefully:

• Gateway unreachable — skips discovery, logs a warning. Manually configured models still work.
• Credits endpoint fails — models still load, but without cost estimates or credit annotations.

License

MIT