kiro-proxy

OpenAI-compatible local proxy for Kiro CLI (Amazon CodeWhisperer / Amazon Q) models.

Expose the free models available through your Kiro login — Claude Opus 4.7, Sonnet 4.6, Haiku 4.5, DeepSeek, Qwen, GLM, Minimax — as a standard OpenAI API, so any OpenAI-compatible client can use them: OpenCode, Cursor, Cline, Kilo Code, Roo Code, Continue, or your own scripts.

Works by reading the OAuth access token that Kiro CLI stores locally after login, and translating between OpenAI chat format and Kiro's internal AmazonCodeWhispererStreamingService.GenerateAssistantResponse protocol.

Quick install (OpenAI-compatible)

One-liner, no git clone needed.

Linux / macOS

curl -fsSL https://raw.githubusercontent.com/bigdata2211it-web/kiro-proxy/main/install/install.sh | bash

Windows (PowerShell)

iwr -useb https://raw.githubusercontent.com/bigdata2211it-web/kiro-proxy/main/install/install.ps1 | iex

The installer:

• checks Node 18+ (prints install hint if missing),
• clones the repo to a sensible OS-specific location,
• runs npm install --omit=dev,
• registers autostart (systemd user unit on Linux, LaunchAgent on macOS, Scheduled Task on Windows),
• smoke-tests http://127.0.0.1:11436.

Details, environment overrides, uninstall: see install/README.md.

---

Features

• ✅ /v1/chat/completions with real streaming (SSE) and non-streaming
• ✅ /v1/models — OpenAI-style model list
• ✅ Tool calling / function calling, full round-trip
• ✅ System prompts, multi-turn history
• ✅ Model name aliases (sonnet → claude-sonnet-4.6, opus → claude-opus-4.7, etc.)
• ✅ Bearer auth read directly from Kiro's local SQLite DB — no re-login, no extra setup

Available models

Model	Context	Rate
auto	1M	1.0x
claude-opus-4.7	1M	2.2x
claude-opus-4.6	1M	2.2x
claude-sonnet-4.6	1M	1.3x
claude-opus-4.5	200K	2.2x
claude-sonnet-4.5	200K	1.3x
claude-sonnet-4	200K	1.3x
claude-haiku-4.5	200K	0.4x
deepseek-3.2	164K	0.25x
minimax-m2.5	196K	0.25x
qwen3-coder-next	256K	0.05x
glm-5	200K	0.5x

Rate = credit multiplier in the Kiro subscription.

Requirements

• Node.js 18+
• Kiro CLI installed and logged in (kiro-cli login once). Get it from kiro.dev.
• Linux, macOS, Windows — all supported. Kiro SQLite path is auto-detected:
  • Linux: $XDG_DATA_HOME/kiro-cli/data.sqlite3 or ~/.local/share/kiro-cli/data.sqlite3
  • macOS: ~/.local/share/kiro-cli/data.sqlite3 (Kiro CLI uses XDG layout)
  • Windows: %APPDATA%\kiro-cli\data.sqlite3
  • Override with KIRO_DB_PATH=/custom/path/data.sqlite3

Install

git clone https://github.com/bigdata2211it-web/kiro-proxy.git
cd kiro-proxy
npm install

Run

Linux / macOS

./start.sh
# or
node index.js

Windows

# PowerShell
.\start.ps1

:: cmd.exe
start.cmd

Listens on http://127.0.0.1:11436. Override port via KIRO_PROXY_PORT.

systemd (Linux, autostart on login)

Create ~/.config/systemd/user/kiro-proxy.service:

[Unit]
Description=Kiro → OpenAI Proxy
After=network.target

[Service]
Type=simple
ExecStart=/usr/bin/node /path/to/kiro-proxy/index.js
Restart=on-failure
RestartSec=5
Environment=KIRO_PROXY_PORT=11436

[Install]
WantedBy=default.target

systemctl --user daemon-reload
systemctl --user enable --now kiro-proxy

Token refresh

Kiro access tokens expire roughly every hour, but any kiro-cli call refreshes them automatically. A simple cron keeps the token fresh:

*/45 * * * * /path/to/kiro-cli chat --no-interactive "ping" >/dev/null 2>&1

Use

OpenCode

In ~/.config/opencode/opencode.json:

{
  "provider": {
    "kiro": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "Kiro (CodeWhisperer)",
      "options": {
        "baseURL": "http://127.0.0.1:11436/v1",
        "apiKey": "dummy"
      },
      "models": {
        "claude-opus-4.7":   { "name": "Claude Opus 4.7 (1M)" },
        "claude-sonnet-4.6": { "name": "Claude Sonnet 4.6 (1M)" },
        "claude-haiku-4.5":  { "name": "Claude Haiku 4.5" },
        "auto":              { "name": "Kiro Auto" }
      }
    }
  },
  "model": "kiro/claude-sonnet-4.6"
}

opencode run --model kiro/claude-opus-4.7 "Refactor this function..."

Cursor / Cline / Kilo Code / Roo Code

Configure an OpenAI-compatible provider:

• Base URL: http://127.0.0.1:11436/v1
• API key: dummy (any string works)
• Model: claude-sonnet-4.6, claude-opus-4.7, auto, …

Continue

In ~/.continue/config.yaml:

models:
  - name: Kiro Sonnet 4.6
    provider: openai
    model: claude-sonnet-4.6
    apiBase: http://127.0.0.1:11436/v1
    apiKey: dummy
    roles: [chat, edit, apply]

curl

# Simple chat
curl http://127.0.0.1:11436/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "claude-sonnet-4.6",
    "messages": [{"role": "user", "content": "Hello"}]
  }'

# With tools
curl http://127.0.0.1:11436/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "claude-sonnet-4.6",
    "messages": [{"role": "user", "content": "list files in /tmp"}],
    "tools": [{
      "type": "function",
      "function": {
        "name": "list_files",
        "description": "List files in a directory",
        "parameters": {
          "type": "object",
          "required": ["path"],
          "properties": {"path": {"type": "string"}}
        }
      }
    }]
  }'

Custom system instructions

The proxy can prepend (or append, or replace) your own system instructions to every request, loaded from plain markdown files. Handy when the client doesn't know about your rules, or you want the same persona/rules across many different clients.

Copy the example config:

mkdir -p ~/.config/kiro-proxy
cp system-instructions.example.json ~/.config/kiro-proxy/system-instructions.json
# edit it, point `files[].path` at your own .md files

Minimal config:

{
  "enabled": true,
  "mode": "prepend",
  "files": [
    { "path": "/abs/path/to/your/AGENTS.md", "required": true }
  ]
}

Modes:

• prepend — your text before whatever the client sent (default).
• append — your text after.
• replace — ignore the client's system prompt entirely.
• off — pass through untouched.

Per-request override: send X-Proxy-Instructions: off (or replace, etc.) with your request to override the configured mode for that call only.

Check what's loaded:

curl -s http://127.0.0.1:11436/debug/instructions | jq .

Force a reload without restart:

curl -s -X POST http://127.0.0.1:11436/debug/instructions/reload | jq .

See examples/ for three ready-to-use packs: minimal/, coding/, persona/.

Your real system-instructions.json and any private rule files are gitignored by default — the repo only ships the example config and the generic example packs.

How it works

OpenAI client
    ↓ POST /v1/chat/completions { messages, tools, stream }
kiro-proxy
    ↓ reads access_token from ~/.local/share/kiro-cli/data.sqlite3
    ↓ converts OpenAI format → Kiro conversationState
    ↓ POST https://q.us-east-1.amazonaws.com/
    ↓ Authorization: Bearer <access_token>
    ↓ X-Amz-Target: AmazonCodeWhispererStreamingService.GenerateAssistantResponse
Kiro / AWS CodeWhisperer
    ↓ AWS Event Stream (binary framed)
kiro-proxy
    ↓ parses events (assistantResponseEvent, toolUseEvent, …)
    ↓ converts to OpenAI chat.completion / SSE
OpenAI client

See PROTOCOL.md for the full Kiro wire protocol.

Limits & known issues

• Images / vision: not implemented (Kiro supports IMAGE input — PR welcome).
• Reasoning / thinking tokens (Opus 4.7 adaptive thinking): not surfaced to clients.
• Rate limits: whatever your Kiro subscription credit pool allows.
• Token refresh: relies on external kiro-cli invocation. Could be replaced by direct OIDC refresh.

Environment variables

Name	Default	Description
KIRO_PROXY_PORT	11436	Listen port.
KIRO_DB_PATH	auto	Override path to Kiro CLI SQLite DB.
KIRO_DUMP_DIR	os.tmpdir()	Where 400-error debug dumps are written.
KIRO_PROXY_INSTRUCTIONS_CONFIG	auto	Path to custom system-instructions.json.

License

MIT. See LICENSE.

Disclaimer

Unofficial. Not affiliated with Amazon, Kiro, or AWS. Use in accordance with your Kiro subscription terms.