Pi

Pi has an OpenAI Responses client built in, openai-responses, and that is the one to point at kiro-provider. A single provider entry carries both the GPT-5.6 and the Claude models.

Add the provider

Pi reads custom endpoints from models.json in its agent directory: ~/.pi/agent/models.json, or the directory PI_CODING_AGENT_DIR names.

{
  "providers": {
    "kiro-provider": {
      "baseUrl": "http://127.0.0.1:8787/v1",
      "api": "openai-responses",
      "apiKey": "$KIRO_GATEWAY_API_KEY",
      "models": [
        {
          "id": "gpt-5.6-sol",
          "name": "GPT-5.6 Sol",
          "reasoning": true,
          "thinkingLevelMap": { "xhigh": "xhigh", "max": "max" },
          "contextWindow": 1000000,
          "maxTokens": 128000,
          "compat": { "supportsMaxOutputTokens": false }
        },
        {
          "id": "claude-opus-5-5",
          "name": "Claude Opus 5.5",
          "reasoning": true,
          "thinkingLevelMap": { "xhigh": "xhigh", "max": "max" },
          "contextWindow": 1000000,
          "maxTokens": 128000
        }
      ]
    }
  }
}

Export the key in the shell that starts Pi, then pick a model on the command line or later with /model:

export KIRO_GATEWAY_API_KEY='<one of your api_keys>'
pi --model kiro-provider/gpt-5.6-sol

What the settings do

Setting	Why
apiKey	$KIRO_GATEWAY_API_KEY makes Pi take the key from the environment variable, so it is not written into the file.
compat	supportsMaxOutputTokens: false. Pi sends max_output_tokens with every request; kiro-provider passes it on to Kiro for claude-opus-5-5, claude-opus-5, claude-sonnet-5 and claude-fable-5-1 only, and refuses it for any other model with unsupported_output_token_limit.
reasoning	Lets /thinking, --thinking or a suffix such as --model kiro-provider/gpt-5.6-sol:high set the reasoning effort.
thinkingLevelMap	Pi offers the xhigh and max levels only for a model that maps them.
contextWindow, maxTokens	The sizes Pi plans with, copied from GET /v1/models. Without them Pi assumes 128,000 and 16,384.

Add other models

GET /v1/models lists every model your accounts can use, with the numbers for contextWindow and maxTokens:

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'

Add each one as another entry under models, with compat.supportsMaxOutputTokens: false unless it is one of the four Claude models above. Names with an effort suffix, such as claude-opus-5-5-high, are not needed: the thinking level sets the effort.

Pi's Anthropic API

Pi's anthropic-messages API reaches the gateway too, at http://127.0.0.1:8787. With reasoning: true, though, Pi asks for summarized thinking, which kiro-provider supports for Claude Fable 5.1 only; other models refuse it with unsupported_reasoning_display. Stay with openai-responses.