Skip to content

AI-Inbox v0.1.3

Latest
Latest

Choose a tag to compare

View all tags
@yujiezhang-ops yujiezhang-ops released this 08 Jul 04:02
· 17 commits to main since this release
v0.1.3
24ff96b

AI-Inbox v0.1.3

AI-Inbox v0.1.3 improves the first-run extraction experience: the app now checks source paths, available session content, LLM protocol, endpoint, API key, model access, and a tiny generation request before running a full organize pass.

Highlights

  • Adds a preflight check before Organize so configuration problems are caught before a full LLM request.
  • Adds LLM protocol selection for OpenAI Chat Completions, OpenAI Responses, and Anthropic Messages.
  • Removes the remaining local token requirement from local workspace write APIs.
  • Makes failed checks readable in the UI, with green pass states and red failure states.

Added

  • ai-inbox preflight [--json] for CLI diagnostics.
  • ai-inbox config get [--json] and ai-inbox config set for the same settings exposed in the UI.
  • AI_INBOX_LLM_PROTOCOL configuration, with legacy AI_INBOX_LLM_PROVIDER=openai mapped to openai-chat.
  • POST /diagnostics/preflight for local UI diagnostics.
  • OpenAI Responses and Anthropic Messages support in the LLM extraction runner.

Changed

  • The default extraction protocol remains OpenAI Chat, with the existing Novita-compatible default model deepseek/deepseek-v4-flash.
  • Settings now keeps advanced LLM fields behind the advanced section, including protocol, endpoint, model, timeout, and API key.
  • Endpoint validation text now says it checks for a valid URL; provider reachability is verified by the model and tiny generation probes.
  • doctor now reports the configured LLM protocol and points users toward ai-inbox preflight for live checks.

Fixed

  • Incorrect API key or provider errors no longer require users to hunt for a diagnostics page; the organize history and settings test show the actionable failure reason directly.
  • Settings and organize write paths no longer fail with a stale local-token Forbidden state.

Compatibility

  • Existing config files continue to work.
  • Existing OpenAI-compatible setups continue to use openai-chat unless AI_INBOX_LLM_PROTOCOL is set.
  • No database schema migration is required for this release.

Downloads

The release zip remains the recommended install path:

  • macOS Apple Silicon: ai-inbox-macos-arm64.zip
  • Windows x64: ai-inbox-windows-x64.zip

Start AI-Inbox with:

ai-inbox start

Then open http://127.0.0.1:3111/.

AI-Inbox v0.1.3 中文说明

AI-Inbox v0.1.3 改善了首次整理前的配置体验：在执行完整整理之前，应用会先检测来源路径、可整理会话内容、LLM 协议、Endpoint、API key、模型访问能力和一次极小的生成请求。

重点变化

  • 点击 整理 前新增预检，先发现配置问题，避免无意义的完整 LLM 请求。
  • 新增 LLM 协议选择：OpenAI Chat Completions、OpenAI Responses、Anthropic Messages。
  • 移除本地工作台写接口中残留的 local token 要求。
  • UI 中的检查结果改为人类可读文案：通过项绿色显示，失败项红色显示。

新增

  • ai-inbox preflight [--json] CLI 诊断命令。
  • ai-inbox config get [--json] 和 ai-inbox config set，覆盖 UI 中可配置的同一批设置。
  • AI_INBOX_LLM_PROTOCOL 配置项；旧的 AI_INBOX_LLM_PROVIDER=openai 会自动映射为 openai-chat。
  • POST /diagnostics/preflight 本地诊断接口。
  • LLM 抽取链路支持 OpenAI Responses 和 Anthropic Messages。

变更

  • 默认抽取协议仍是 OpenAI Chat，默认模型仍是 deepseek/deepseek-v4-flash。
  • Settings 中的协议、Endpoint、模型、timeout、API key 等仍放在高级 LLM 配置区。
  • Endpoint 检测文案明确为“有效 URL”；provider 是否可用由模型列表和极小生成请求继续验证。
  • doctor 现在会显示 LLM protocol，并提示使用 ai-inbox preflight 做真实链路检测。

修复

  • API key 或 provider 错误会直接显示在整理历史和设置测试结果里，不再引导用户寻找不存在的诊断页。
  • Settings 保存、整理等本地写操作不再因为旧 local token 状态返回 Forbidden。

兼容性

  • 现有配置文件继续可用。
  • 现有 OpenAI-compatible 配置默认继续走 openai-chat，除非显式设置 AI_INBOX_LLM_PROTOCOL。
  • 本版本不需要数据库 schema 迁移。

下载

推荐继续使用 release zip：

  • macOS Apple Silicon：ai-inbox-macos-arm64.zip
  • Windows x64：ai-inbox-windows-x64.zip

启动：

ai-inbox start

然后打开 http://127.0.0.1:3111/。

Assets 5
Loading