OpenCode

OpenCode installs the @ai-sdk/anthropic package on its own, so kiro-provider joins it as an Anthropic Messages provider. One entry covers the Claude and the GPT-5.6 models.

Add the provider

Put the provider in ~/.config/opencode/opencode.json, or in opencode.json at the root of a project:

{
  "$schema": "https://opencode.ai/config.json",
  "provider": {
    "kiro-provider": {
      "npm": "@ai-sdk/anthropic",
      "name": "kiro-provider",
      "options": {
        "baseURL": "http://127.0.0.1:8787/v1",
        "apiKey": "{env:KIRO_GATEWAY_API_KEY}",
        "headers": {
          "x-kiro-output-token-limit-mode": "advisory"
        }
      },
      "models": {
        "claude-opus-5-5": {
          "name": "Claude Opus 5.5",
          "limit": { "context": 1000000, "output": 128000 },
          "options": { "toolStreaming": false }
        },
        "gpt-5.6-sol": {
          "name": "GPT-5.6 Sol",
          "limit": { "context": 1000000, "output": 128000 },
          "options": { "thinking": { "type": "enabled", "budgetTokens": 16000 } }
        }
      }
    }
  },
  "model": "kiro-provider/claude-opus-5-5"
}

Export the key in the shell that starts OpenCode. /models switches between the two models, and -m picks one for a single run:

export KIRO_GATEWAY_API_KEY='<one of your api_keys>'
opencode
opencode run -m kiro-provider/gpt-5.6-sol "Reply with exactly: KIRO_OK"

What the settings do

Setting	Why
baseURL	Ends in /v1: @ai-sdk/anthropic adds /messages to it, where the Anthropic SDKs add /v1/messages.
headers	OpenCode sends max_tokens with every request. Kiro has no output limit for the GPT-5.6 models, so kiro-provider refuses their requests unless x-kiro-output-token-limit-mode: advisory is present. With it, kiro-provider leaves the limit out for those models only.
toolStreaming	false on Claude models. Otherwise OpenCode marks their tools with eager_input_streaming, which kiro-provider refuses.
thinking	On for GPT-5.6 models. GPT-5.6 can send its reasoning signature after the answer has begun. With thinking on, kiro-provider waits for it; without, such a reply fails with Upstream returned invalid reasoning metadata.
limit	The context and output sizes OpenCode plans with, copied from GET /v1/models.

Which models work

Every Messages request carries max_tokens. kiro-provider passes it on to Kiro for four Claude models: claude-opus-5-5, claude-opus-5, claude-sonnet-5 and claude-fable-5-1, from 1,024 to 128,000. With the header above, gpt-5.6-sol, gpt-5.6-terra and gpt-5.6-luna work too. Any other model is refused with an invalid_request_error about max_tokens; use it from a Responses client such as Pi.

Add a model under models with the numbers GET /v1/models reports, toolStreaming: false for a Claude model and thinking for a GPT-5.6 model:

curl -s http://127.0.0.1:8787/v1/models -H "Authorization: Bearer $KIRO_GATEWAY_API_KEY" |
  jq -r '.data[] | [.id, .context_limit, .output_limit] | @tsv'

The Responses provider

OpenCode can also reach the gateway with @ai-sdk/openai at the same baseURL. Claude models work that way; GPT-5.6 does not. Through that package OpenCode sends max_output_tokens on every request, and the advisory header applies to Messages only, so a GPT-5.6 request fails with unsupported_output_token_limit. The Messages setup above covers both.

Accounts from opencode-kiro-auth

If you have been using Kiro in OpenCode through the opencode-kiro-auth plugin, copy those accounts into kiro-provider once and stop using the plugin's provider: two programs renewing one refresh token race each other. Accounts covers the import.