feat: upstream providers[] + convert_from_map (#131)

* feat: upstream providers[] + convert_from_map

- Replace upstream.provider with providers[]

- Add per-upstream convert_from_map for cross-format fallback

- Add config migration + UI editor updates

* test(frontend): add Vitest to prevent config/dashboard regressions

So config normalization, upstream defaults, and dashboard helpers stay reliable
as the UI evolves and CI runs.

* perf: cut overhead when tracking or capture is disabled

Avoid blocking Tokio threads on temp-file cleanup and reduce
allocations/background tasks for noop token tracking; also skip
redundant log capture-state loads.

Author: lightlifedev

README.md

@@ -9,7 +9,7 @@ Local AI API gateway for OpenAI / Gemini / Anthropic. Runs on your machine, keep
 ---
 
 ## What you get
-- Multiple providers: `openai`, `openai-response`, `anthropic`, `gemini`, `kiro`
+- Multiple providers: `openai`, `openai-response`, `anthropic`, `gemini`, `kiro`, `codex`, `antigravity`
 - Built-in routing + optional format conversion (OpenAI Chat ⇄ Responses; Anthropic Messages ↔ OpenAI; Gemini ↔ OpenAI/Anthropic; SSE supported)
 - Per-upstream priority + two balancing strategies (fill-first / round-robin)
 - Model alias mapping (exact / prefix* / wildcard*) and response model rewrite
@@ -62,6 +62,26 @@ cargo run -p token_proxy_cli -- config init
 cargo run -p token_proxy_cli -- --config ./config.jsonc config path
 ```
 
+## Frontend tests
+```bash
+# watch mode
+pnpm test
+
+# run once (CI-friendly)
+pnpm test:run
+
+# coverage (optional)
+pnpm test:coverage
+
+# TypeScript typecheck
+pnpm exec tsc --noEmit
+```
+
+Notes:
+- Test files live in `src/**/*.test.{ts,tsx}`.
+- Global test setup (Tauri mocks + jsdom polyfills) is in `src/test/setup.ts`.
+- Vitest config is in `vitest.config.ts`.
+
 ## Configuration reference
 - File: `config.jsonc` (comments + trailing commas allowed)
 - Location:
@@ -79,33 +99,33 @@ cargo run -p token_proxy_cli -- --config ./config.jsonc config path
 | `max_request_body_bytes` | `20971520` (20 MiB) | 0 = fallback to default. Protects inbound body size. |
 | `tray_token_rate.enabled` | `true` | macOS tray live rate; harmless elsewhere. |
 | `tray_token_rate.format` | `split` | `combined` (`total`), `split` (`↑in ↓out`), `both` (`total | ↑in ↓out`). |
-| `enable_api_format_conversion` | `true` | Allow OpenAI/Anthropic/Gemini fallback via request/response body and SSE stream conversion. |
 | `upstream_strategy` | `priority_fill_first` | `priority_fill_first` (default) keeps trying the highest-priority group in list order; `priority_round_robin` rotates within each priority group. |
 
 ### Upstream entries (`upstreams[]`)
 | Field | Default | Notes |
 | --- | --- | --- |
 | `id` | required | Unique per upstream. |
-| `provider` | required | One of `openai`, `openai-response`, `anthropic`, `gemini`, `kiro`. |
-| `base_url` | required | Full base; overlapping path parts are de-duplicated. (`kiro` can be empty.) |
+| `providers` | required | One upstream can serve multiple providers. Special providers `kiro/codex/antigravity` cannot be mixed with others. |
+| `base_url` | required | Full base; overlapping path parts are de-duplicated. (`providers=["kiro"]` / `["codex"]` / `["antigravity"]` can be empty.) |
 | `api_key` | `null` | Provider-specific bearer/key; overrides request headers. |
-| `kiro_account_id` | `null` | Required when `provider=kiro`. |
-| `preferred_endpoint` | `null` | `kiro` only: `ide` or `cli`. |
+| `kiro_account_id` | `null` | Required when `providers=["kiro"]`. |
+| `preferred_endpoint` | `null` | `kiro` only (`providers=["kiro"]`): `ide` or `cli`. |
 | `proxy_url` | `null` | Per-upstream proxy; supports `http/https/socks5/socks5h`; default is **no system proxy**. `$app_proxy_url` placeholder allowed. |
 | `priority` | `0` | Higher = tried earlier. Grouped by priority then by order (or round-robin). |
 | `enabled` | `true` | Disabled upstreams are skipped. |
 | `model_mappings` | `{}` | Exact / `prefix*` / `*`. Priority: exact > longest prefix > wildcard. Response echoes original alias. |
+| `convert_from_map` | `{}` | Explicitly allow inbound format conversion per provider. Example: `{ "openai-response": ["openai_chat", "anthropic_messages"] }`. |
 | `overrides.header` | `{}` | Set/remove headers (null removes). Hop-by-hop/Host/Content-Length are always ignored. |
 
 ## Routing & format conversion
 - Gemini: `/v1beta/models/*:generateContent` and `*:streamGenerateContent` → `gemini` (SSE supported).
 - Anthropic: `/v1/messages` (and subpaths) and `/v1/complete` → `anthropic` (Kiro shares the same format).
 - OpenAI: `/v1/chat/completions` → `openai`; `/v1/responses` → `openai-response`.
 - Other paths: choose the provider with the highest configured priority; tie-break is `openai` > `openai-response` > `anthropic`.
-- If the preferred provider is missing but `enable_api_format_conversion=true`, the proxy auto-converts request/response bodies and streams between supported formats (including SSE).
+- Cross-format fallback/conversion is controlled by `upstreams[].convert_from_map` (no global switch). If a provider has no eligible upstream for the inbound format, it won't be selected.
 - If `openai` is missing for `/v1/chat/completions`: fallback can be `openai-response`, `anthropic`, or `gemini` (priority-based; tie-break prefers `openai-response`).
 - For `/v1/messages`: choose between `anthropic` and `kiro` by priority; tie-break uses upstream id. If the chosen provider returns a retryable error, the proxy will fall back to the other native provider (Anthropic ↔ Kiro) when configured.
-- If neither `anthropic` nor `kiro` exists for `/v1/messages` and `enable_api_format_conversion=true`: fallback can be `openai-response`, `openai`, or `gemini` (priority-based; tie-break prefers `openai-response`).
+- If neither `anthropic` nor `kiro` exists for `/v1/messages`: fallback can be `openai-response`, `openai`, or `gemini` when the target provider is allowed for `anthropic_messages` via `convert_from_map`.
 - If `openai-response` is missing for `/v1/responses`: fallback can be `openai`, `anthropic`, or `gemini` (priority-based; tie-break prefers `openai`).
 - If `gemini` is missing for `/v1beta/models/*:generateContent`: fallback can be `openai-response`, `openai`, or `anthropic` (priority-based; tie-break prefers `openai-response`).
 

README.zh-CN.md

@@ -9,7 +9,7 @@
 ---
 
 ## 你能得到什么
-- 多提供商：`openai`、`openai-response`、`anthropic`、`gemini`、`kiro`
+- 多提供商：`openai`、`openai-response`、`anthropic`、`gemini`、`kiro`、`codex`、`antigravity`
 - 内置路由，支持可选的 API 格式互转（OpenAI Chat ⇄ Responses；Anthropic Messages ↔ OpenAI；Gemini ↔ OpenAI/Anthropic，含 SSE）
 - 上游优先级 + 两种策略（填满优先级组 / 轮询）
 - 模型别名映射（精确 / 前缀* / 通配*），响应会回写原始别名
@@ -62,6 +62,26 @@ cargo run -p token_proxy_cli -- config init
 cargo run -p token_proxy_cli -- --config ./config.jsonc config path
 ```
 
+## 前端测试
+```bash
+# watch 模式
+pnpm test
+
+# 单次运行（CI 友好）
+pnpm test:run
+
+# 覆盖率（可选）
+pnpm test:coverage
+
+# TypeScript 类型检查
+pnpm exec tsc --noEmit
+```
+
+说明：
+- 测试文件约定：`src/**/*.test.{ts,tsx}`。
+- 全局测试初始化（Tauri mocks + jsdom polyfills）：`src/test/setup.ts`。
+- Vitest 配置：`vitest.config.ts`。
+
 ## 配置参考
 - 文件：`config.jsonc`（支持注释与尾随逗号）
 - 位置：
@@ -79,33 +99,33 @@ cargo run -p token_proxy_cli -- --config ./config.jsonc config path
 | `max_request_body_bytes` | `20971520` (20 MiB) | 0 表示回落到默认；保护入站体积 |
 | `tray_token_rate.enabled` | `true` | macOS 托盘实时速率；其他平台无害 |
 | `tray_token_rate.format` | `split` | `combined`(总数) / `split`(↑入 ↓出) / `both`(总数 | ↑入 ↓出) |
-| `enable_api_format_conversion` | `true` | 允许 OpenAI/Anthropic/Gemini 自动 fallback（含请求/响应体转换与 SSE 流式转换） |
 | `upstream_strategy` | `priority_fill_first` | `priority_fill_first` 默认先填满高优先级；`priority_round_robin` 在同组内轮询 |
 
 ### 上游条目（`upstreams[]`）
 | 字段 | 默认值 | 说明 |
 | --- | --- | --- |
 | `id` | 必填 | 唯一 |
-| `provider` | 必填 | `openai` / `openai-response` / `anthropic` / `gemini` / `kiro` |
-| `base_url` | 必填 | 完整基址，重复路径段会去重（`kiro` 可为空） |
+| `providers` | 必填 | 一个上游可同时服务多个 provider。特殊 provider（`kiro/codex/antigravity`）不可与其它 provider 混用。 |
+| `base_url` | 必填 | 完整基址，重复路径段会去重（`providers=["kiro"]` / `["codex"]` / `["antigravity"]` 可为空） |
 | `api_key` | `null` | 该 provider 的密钥；优先于请求头 |
-| `kiro_account_id` | `null` | `provider=kiro` 时必填 |
+| `kiro_account_id` | `null` | `providers=["kiro"]` 时必填 |
 | `preferred_endpoint` | `null` | `kiro` 专用：`ide` 或 `cli` |
 | `proxy_url` | `null` | 每个上游独立代理，支持 `http/https/socks5/socks5h`；默认**不走系统代理**；支持 `$app_proxy_url` |
 | `priority` | `0` | 越大越先尝试；同组按列表顺序或轮询 |
 | `enabled` | `true` | 可临时禁用上游 |
 | `model_mappings` | `{}` | 精确 / `前缀*` / `*`；优先级：精确 > 最长前缀 > 通配；响应回写原始模型别名 |
+| `convert_from_map` | `{}` | 显式声明允许从哪些入站格式转换后使用该 provider。例：`{ "openai-response": ["openai_chat", "anthropic_messages"] }` |
 | `overrides.header` | `{}` | 设置/删除 header（null 表示删除）；hop-by-hop/Host/Content-Length 永远忽略 |
 
 ## 路由与格式转换
 - Gemini：`/v1beta/models/*:generateContent`、`*:streamGenerateContent` → `gemini`（支持 SSE）
 - Anthropic：`/v1/messages`（含子路径）与 `/v1/complete` → `anthropic`（Kiro 同格式）
 - OpenAI：`/v1/chat/completions` → `openai`；`/v1/responses` → `openai-response`
 - 其他路径：按已配置 provider 的最高优先级选择；优先级相同则按 `openai` > `openai-response` > `anthropic` 打破平局
-- 若首选 provider 缺失且 `enable_api_format_conversion=true`，将自动在已支持的格式之间转换请求与响应（含 SSE 流式）
+- 跨格式 fallback/转换由 `upstreams[].convert_from_map` 控制（不再有全局开关）；若某个 provider 在该入站格式下没有任何可用 upstream，则不会被选中。
 - `/v1/chat/completions` 缺少 `openai`：可 fallback 到 `openai-response` / `anthropic` / `gemini`（按优先级选择，平级优先 `openai-response`）
 - `/v1/messages`：在 `anthropic` 与 `kiro` 间按优先级选择；平级按 upstream id 排序。若命中 provider 返回“可重试错误”，且另一个 native provider 已配置，则会自动 fallback（Anthropic ↔ Kiro）
-- 当 `/v1/messages` 缺少 `anthropic` 且 `kiro` 也不存在，且 `enable_api_format_conversion=true` 时：可 fallback 到 `openai-response` / `openai` / `gemini`（按优先级选择，平级优先 `openai-response`）
+- 当 `/v1/messages` 缺少 `anthropic` 且 `kiro` 也不存在时：若目标 provider 在 `convert_from_map` 中允许 `anthropic_messages`，则可 fallback 到 `openai-response` / `openai` / `gemini`（按优先级选择，平级优先 `openai-response`）
 - `/v1/responses` 缺少 `openai-response`：可 fallback 到 `openai` / `anthropic` / `gemini`（按优先级选择，平级优先 `openai`）
 - `/v1beta/models/*:generateContent` 缺少 `gemini`：可 fallback 到 `openai-response` / `openai` / `anthropic`（按优先级选择，平级优先 `openai-response`）
 

crates/token_proxy_core/src/proxy/config/io.rs

@@ -4,14 +4,23 @@ use std::time::Instant;
 use crate::paths::TokenProxyPaths;
 
 use super::ProxyConfigFile;
+use super::migrate::migrate_config_json;
 
 const DEFAULT_CONFIG_HEADER: &str = concat!(
     "// Token Proxy config (JSONC). Comments and trailing commas are supported.\n",
     "// log_level (optional): silent|error|warn|info|debug|trace. Default: silent.\n",
     "// app_proxy_url (optional): http(s)://... | socks5(h)://... (used for app updates and upstream proxy reuse).\n",
-    "// upstreams[].proxy_url (optional): empty => direct; \"$app_proxy_url\" => use app_proxy_url; or an explicit proxy URL.\n"
+    "// upstreams[].proxy_url (optional): empty => direct; \"$app_proxy_url\" => use app_proxy_url; or an explicit proxy URL.\n",
+    "// upstreams[].providers (required): one upstream can serve multiple providers. Example: [\"openai\", \"openai-response\"].\n",
+    "// upstreams[].convert_from_map (optional): explicitly allow inbound format conversion per provider.\n",
+    "//   Example: { \"openai-response\": [\"openai_chat\", \"anthropic_messages\"] }\n"
 );
 
+struct ParsedConfigFile {
+    config: ProxyConfigFile,
+    migrated: bool,
+}
+
 pub(super) async fn load_config_file(paths: &TokenProxyPaths) -> Result<ProxyConfigFile, String> {
     let path = paths.config_file();
     tracing::debug!(path = %path.display(), "load_config_file start");
@@ -24,7 +33,12 @@ pub(super) async fn load_config_file(paths: &TokenProxyPaths) -> Result<ProxyCon
                 elapsed_ms = start.elapsed().as_millis(),
                 "load_config_file read"
             );
-            parse_config_file(&contents, &path)
+            let parsed = parse_config_file(&contents, &path)?;
+            if parsed.migrated {
+                tracing::info!(path = %path.display(), "config migrated, writing back");
+                save_config_file(paths, &parsed.config).await?;
+            }
+            Ok(parsed.config)
         }
         Err(err) if err.kind() == std::io::ErrorKind::NotFound => {
             tracing::debug!(
@@ -91,10 +105,14 @@ pub(super) async fn init_default_config_file(paths: &TokenProxyPaths) -> Result<
     save_config_file(paths, &ProxyConfigFile::default()).await
 }
 
-fn parse_config_file(contents: &str, path: &Path) -> Result<ProxyConfigFile, String> {
+fn parse_config_file(contents: &str, path: &Path) -> Result<ParsedConfigFile, String> {
     let sanitized = crate::jsonc::sanitize_jsonc(contents);
-    serde_json::from_str(&sanitized)
-        .map_err(|err| format!("Failed to parse config file {}: {err}", path.display()))
+    let mut value: serde_json::Value = serde_json::from_str(&sanitized)
+        .map_err(|err| format!("Failed to parse config file {}: {err}", path.display()))?;
+    let migrated = migrate_config_json(&mut value);
+    let config: ProxyConfigFile = serde_json::from_value(value)
+        .map_err(|err| format!("Failed to parse config file {}: {err}", path.display()))?;
+    Ok(ParsedConfigFile { config, migrated })
 }
 
 async fn read_existing_header(path: &Path) -> Option<String> {