Repository navigation
Releases
Releases · ProTechPh/Kiro-account-manager
Kiro Account Manager v1.8.1
🌟 更新内容 | What's New
v1.8.1
新功能：
- ⚡ 开机自启动: 系统启动时自动运行应用（原生系统登录项）
- 🚀 自动启动服务器: 应用打开时自动启动代理服务器（设置页面开关）
- 🛠️ 自动修复: 启动/重启时连通性自检 — 探测
/v1/models，401 时自动轮换 API Key，服务无响应时自动重启
详见: docs/CHANGELOG-v1.8.1.md
v1.8.1
New Features:
- ⚡ Launch at Startup: Automatically start the app when your system boots (native OS login item)
- 🚀 Auto-Start Server: Automatically start the proxy server when the app opens (Settings toggle)
- 🛠️ Auto Repair: Connectivity self-check on start/restart — probes
/v1/models, rotates API key on 401, restarts on unresponsive service
See full details: docs/CHANGELOG-v1.8.1.md
Kiro Account Manager v1.8.0
🌟 更新内容 | What's New
v1.8.0
新功能：
- 🌐 公网访问（Cloudflare Tunnel）: 一键通过
cloudflared快速隧道生成公网 URL — 免费 HTTPS，无需账号 - 🔀 代理池: 出站代理管理，支持 HTTP/SOCKS、Vercel 中继部署、批量导入
- 📊 实时配额: 实时 Token 追踪、费用估算、每账号配额进度条及重置倒计时
- 🛡️ MITM 桥接: 在端口 443 拦截 IDE HTTPS 流量，根 CA 生成、动态 SNI 证书、请求转发
详见: docs/CHANGELOG-v1.8.0.md
v1.8.0
New Features:
- 🌐 Public Access (Cloudflare Tunnel): One-click public URL via
cloudflaredquick tunnel — free HTTPS, no account needed - 🔀 Proxy Pools: Outbound proxy management with HTTP/SOCKS support, Vercel Relay deployment, batch import
- 📊 Real-Time Quota: Live token tracking, cost estimation, per-account quota progress bars with reset countdown
- 🛡️ MITM Bridge: Intercept IDE HTTPS traffic on port 443 with Root CA generation, dynamic SNI certs, and request forwarding
See full details: docs/CHANGELOG-v1.8.0.md
Kiro Account Manager v1.7.1
🌟 更新内容 | What's New
v1.7.1
新功能:
- 🛡️ 熔断器指数退避: 智能冷却
base × 2^(failures-1)+ 概率性重试（10%），防止账号永久卡死 - ⏱️ 可配置流式超时: 首 Token 超时（15s）、最大重试（3次）、流式读取超时（5分钟）— 全部可在 UI 调整
- 🔍 Web Search 工具自动注入: MCP 模拟 — 自动注入
web_search工具让模型可以搜索网络 - 📦 Payload 大小保护: 可配置限制（默认 600KB）+ 自动裁剪开关 — 不再静默丢失数据
- 🧠 思考预算上限: 默认预算（4000 tokens）+ 最大上限（10000），防止模型将所有输出花在推理上
- 🖥️ 10 个新 UI 设置项: 所有功能均在高级设置面板中暴露，完整中英文翻译
完整说明见: docs/CHANGELOG-v1.7.1.md
v1.7.1
New Features:
- 🛡️ Circuit Breaker Exponential Backoff: Smart cooldown with
base × 2^(failures-1)formula + probabilistic retry (10% chance) to prevent permanent stuck accounts - ⏱️ Configurable Streaming Timeouts: First token timeout (15s), max retries (3), streaming read timeout (5min) — all adjustable from UI
- 🔍 Web Search Tool Auto-Injection: MCP emulation — auto-inject
web_searchtool so models can search the web - 📦 Payload Size Guard: Configurable limit (600KB default) with auto-trim toggle — no more silent data loss
- 🧠 Fake Reasoning Budget Cap: Default budget (4000 tokens) + max cap (10000) to prevent model spending all output on thinking
- 🖥️ 10 New UI Settings: All features exposed in the Advanced Settings panel with full EN/ZH translations
See full details: docs/CHANGELOG-v1.7.1.md
Kiro Account Manager v1.7.0
🌟 更新内容 | What's New
Kiro Account Manager v1.6.9
🌟 更新内容 | What's New
v1.6.9
Bug 修复:
- 🐛 AI 响应中回显占位符文本: 修复了 AI 模型将助手历史消息中的任意占位符文本（如
"Using tools."、"(tool_call)"）原样回显到实际响应中的问题。现已彻底移除包含工具调用的助手消息中的所有占位符内容。
完整说明见: docs/CHANGELOG-v1.6.9.md
v1.6.9
Bug Fixes:
- 🐛 Placeholder Text Echo in AI Responses: Fixed a persistent issue where ANY placeholder text in assistant history messages (e.g.
"Using tools.","(tool_call)") was being echoed by the AI model into actual responses. Removed all placeholder content from tool-call assistant messages entirely.
See full details: docs/CHANGELOG-v1.6.9.md
Kiro Account Manager v1.6.8
🌟 更新内容 | What's New
v1.6.8
Bug 修复:
- 🐛 OpenAI 兼容模式响应中出现 "Using tools." 字样: 修复了在使用 OpenAI 兼容代理时，内部占位字符串
"Using tools."意外出现在 AI 实际响应中的问题。该字符串是代理为满足 Kiro API 对历史消息assistantResponseMessage.content非空的要求而注入的，但模型会将其视为正常对话内容并在下一轮中回显。现已替换为(tool_call)，这是一个括号式元数据标记，模型能正确将其识别为非对话内容而不予复述。
完整说明见: docs/CHANGELOG-v1.6.8.md
v1.6.8
Bug Fixes:
- 🐛 OpenAI-Compatible "Using tools." Leak: Fixed a bug where the internal placeholder string
"Using tools."was appearing in actual AI responses when using the OpenAI-compatible proxy. The string—required by the Kiro API to satisfy a non-empty history constraint—was being echoed back by the model. Replaced with(tool_call), a parenthetical metadata marker the model correctly ignores.
See full details: docs/CHANGELOG-v1.6.8.md
Kiro Account Manager v1.6.7
🌟 更新内容 | What's New
v1.6.7
Bug 修复:
- 🐛 思考阶段 Token 耗尽瘫痪: 修复了由于
max_thinking_length上限被设为 200,000 导致 AI 在深度思考时完全榨干 8,192 最大输出 Token，最终造成返回流严重截断崩溃的问题。现将上限限制至 4,000 tokens，以安全保留容量用于正常的文件代码及终端打印。
完整说明见: docs/CHANGELOG-v1.6.7.md
v1.6.7
Bug Fixes:
- 🐛 Thinking Token Exhaustion: Fixed a prompt truncation issue by capping the proxy's
max_thinking_lengthto 4,000 tokens instead of 200,000, reserving adequate capacity for the actual file writes and tool calls without blowing past the 8,192 API limits limit.
See full details: docs/CHANGELOG-v1.6.7.md
Kiro Account Manager v1.6.6
🌟 更新内容 | What's New
v1.6.6
Bug 修复:
- 🐛 工具调用超时崩溃: 修复了巨大的 IDE Agent 工具流（如超大文件写入）导致错误触发 15 秒首 Token 超时的问题。现在会在解析到第一个有效网络包时立即停止初次等待计时器以防无故中断。
- 🐛 丢失的流异常: 解决了内部
parseEventStream工具流解析异常被吞没导致连接无限挂起的问题，让异常能准确冒泡至重试连接池以触发账号平滑自动切换轮换。
完整说明见: docs/CHANGELOG-v1.6.6.md
v1.6.6
Bug Fixes:
- 🐛 Tool Call Execution Timeouts: Fixed a critical bug where massive IDE agent tool use requests would trip the proxy's 15-second first-token timeout before full completion. We now proactively signal the proxy as soon as the initial data packet arrives.
- 🐛 Lost Stream AbortErrors: Solved an infinite hang scenario where internal runtime token chunking errors were swallowed instead of passing up to the core connection pool. The proxy now properly transitions to the retry handler for seamless endpoint recovery.
See full details: docs/CHANGELOG-v1.6.6.md
Kiro Account Manager v1.6.5
🌟 更新内容 | What's New
v1.6.4
Bug 修复:
- 🐛 API 配额耗尽与智能重试机制: 修复了在遇到通用速率限制 (
429) 时引发的 API 无限重试循环问题。现在针对服务器瞬时高并发情况，系统使用指数退避算法 (1s,2s,4s延迟) 智能轮询端点；并在遭遇绝对配额耗尽 (Monthly limit exceeded) 时立刻阻断，安全交由AccountPool无缝进行账号自动切换。
完整说明见: docs/CHANGELOG-v1.6.4.md
v1.6.4
Bug Fixes:
- 🐛 API Quota Exhaustion & Smart Retry: Fixed an infinite API retry loop on generic rate limits (
429). Now intelligently prioritizes fallback endpoints with exponential backoff (1s,2s,4sdelays) for transient high volume, while fully enabling theAccountPoolto auto-switch instances instantaneously upon hard limits (Monthly limit exceeded).
See full details: docs/CHANGELOG-v1.6.4.md
Kiro Account Manager v1.6.3
🌟 更新内容 | What's New
v1.6.3
Bug 修复:
- 🐛 多主体 (Multi-Agent) 流稳定性: 禁用了系统在流结束后启动的 8 秒请求阻断期。这修复了类似 Claude Code 在多主体自动化任务中无法发送后续请求的问题，避免了被误认为"用户手动取消" (HTTP 499 错误)。
完整说明见: docs/CHANGELOG-v1.6.3.md
v1.6.3
Bug Fixes:
- 🐛 Multi-Agent Streaming Stability: Disabled the 8-second request blockage after a stream ends. This fixes the issue preventing tools like Claude Code from properly establishing subsequent requests in multi-agent workflows, avoiding false "user stop" cancellations (HTTP 499).