@goodandready/dsh-opencode-provider 0.5.21 → 0.5.23

package/CHANGELOG.md

@@ -1,5 +1,20 @@
 # Changelog
 
+## 0.5.23
+
+- **DSH 0.2.2 support**: compatibility now accepts every `0.2.2` prerelease and final release with `>=0.2.1-rc.2 <0.2.3-0`; `0.2.2-rc.1` is recorded as verified. Dev/typecheck dependencies moved to `@deepseek-ai/*@0.2.2-rc.1`, Cordis 4.1.0 and Schemastery 3.19.0; peer ranges admit both the 0.2.1 and 0.2.2 lines.
+- **Host webserver compatibility**: route registration now types handlers against `WebRoute` instead of indexing the new `WebRoute | WebResourceRoute` union, preserving 0.2.1 behavior while compiling on 0.2.2.
+- **Verification**: isolated DSH 0.2.2-rc.1 web profile activated both routes, discovered 29 models and completed a real `deepseek-v4.1-flash` request (`OK`); 0.2.1-rc.4 and 0.2.1-rc.2 activation regressions also pass. `npm run check`: 124/124 tests.
+
+## 0.5.22
+
+- **Proxy transport**: one live `proxyUrl` setting (HTTP, HTTPS or SOCKS5) now covers inference, catalog discovery and usage checks. Dispatchers are shared by address and released when the plugin unloads; proxy credentials are rejected so the settings API never returns a secret.
+- **Account pool parity**: up to 20 accounts, drag-and-drop priority in Settings, promotion of the account that served the last request (persisted alongside sticky state), and a 5-minute invalid-credential cooldown matching the reference plugin.
+- **Per-model overrides**: per-model context window and output caps are now editable per model; `0` omits the output cap. The output override also applies to request construction, not only the model descriptor.
+- **Tool-call validation**: numeric/null ids and names from OpenAI Chat, Anthropic, Responses and Google streams are normalized to string fallbacks before emitting `tool-call-delta`; a repeated `id:null` no longer overwrites the id from the first fragment. Six regression tests cover streaming, continuation, non-SSE and all transports (GitHub #2).
+- **Docs/TUI**: README/RU/ZH mirrors and the TUI connection section expose the same fields; 124 tests.
+- **Not added**: a separate keyless Free route. A live probe returned HTTP 429 `FreeUsageLimitError`; the existing keyed Console/Zen route remains the reliable path.
+
 ## 0.5.21
 
 - **Analytics completeness**: projection scans above 5000 files now select the newest files by metadata instead of an arbitrary readdir page; file-limit and unreadable-projection cases set `AnalyticsReport.partial`, surfaced in the dashboard.

package/README.md

@@ -91,16 +91,18 @@ graph LR
 - **Replay of provider metadata** where the endpoint requires it:
   - Anthropic thinking signatures.
   - Gemini `thoughtSignature` on historical function calls.
-- **Multi-account rotation** before a stream starts: 401/403/429 and quota/usage-limit bodies (400/402/403/429). An invalid key cools down for 24h, a quota rejection for 10 minutes or `Retry-After`, a rate limit for 60s or `Retry-After`. A request already streaming is never replayed on another account; the next request uses the next usable key.
+- **Multi-account rotation** before a stream starts: up to 20 accounts in drag-and-drop priority order, 401/403/429 and quota/usage-limit bodies (400/402/403/429). An invalid key cools down for 5 minutes, a quota rejection for 10 minutes or `Retry-After`, a rate limit for 60s or `Retry-After`. The account that served the last request is promoted for following conversations and survives restarts. A request already streaming is never replayed on another account; the next request uses the next usable key.
 - **Per-account quotas**: each account row shows its own 5-hour / weekly / monthly usage, the active account drives the sidebar and quota panel, and the panel lists every account side by side.
 - **Per-account runway**: estimates time to exhaustion from the average burn rate of each window and marks the first window projected to run out before its reset; approximate because the API exposes only the current percentage and reset time.
 - **Model → account rules**: first matching glob pins a preferred account for that model; the other accounts remain fallbacks if it fails or is cooling.
+- **Per-model overrides**: context window and output cap can be changed per model in Settings; `0` omits the output cap for that model.
 - **Request heatmap** in the quota panel: a Monday-first weekday × hour grid in local time, plus a 24-hour aggregate strip underneath. Selectors switch between all requests, any single model, and any single account. A per-account table shows requests, errors and average latency/TTFT. The adapter appends `{at, provider, model, purpose?, account, reason, ok, latencyMs, ttftMs?, error?}` to `~/.dsh/storages/dsh-opencode/requests.jsonl`; no prompt, response or key is stored. Auxiliary `session-title`/`compaction` calls are excluded from the heatmap and exact duplicate records collapse. The log compacts at 2 MB to the newest 5000 lines. Collection starts with this build — the projection cache only has `lastPromptAt`, not per-request timestamps.
 - **Stable `x-opencode-session`** derived from the durable session id, so retries and every turn of a conversation reuse the same routing key.
 - **Live catalog** from `/v1/models` with a generated `models.dev` fallback.
+- **Proxy support**: one live `proxyUrl` setting routes inference, model discovery and usage checks through HTTP, HTTPS or SOCKS5; the dispatcher is reused until the address changes.
 - **Per-model subscription limits** from the OpenCode Go pricing table: monthly dollar allowance plus estimated requests for the 5-hour, weekly and monthly windows, for both Go and Go Plus. Shown in the model picker and the settings page; estimates only — the gateway remains the authority.
 - **`/opencode` command** for status, usage, accounts, models, rules, and `test [model]`.
-- **TUI settings section** for the terminal profile (`/settings → OpenCode`) with the key, endpoints, allowlist, active account, fallback model, sticky routing and timeouts.
+- **TUI settings section** for the terminal profile (`/settings → OpenCode`) with the key, endpoints, proxy, allowlist, active account, fallback model, sticky routing and timeouts.
 - **Quota fallback**: choose a free model in Settings → OpenCode; a Go quota rejection retries once through Console/Zen.
 - **Sticky account**: the last successful account is preferred for the same conversation and survives restarts via a local `sticky.json` (hashed session ids and key fingerprints only); explicit model rules and pinned accounts win.
 - **Routing trace** in the quota panel and per-account heatmap filter.
@@ -130,12 +132,12 @@ npm pack
 Install the **packed tarball** into a profile:
 
 ```bash
-dsh plugin --profile web add /absolute/path/to/goodandready-dsh-opencode-provider-0.5.21.tgz
+dsh plugin --profile web add /absolute/path/to/goodandready-dsh-opencode-provider-0.5.23.tgz
 # DSH Desktop:
-dsh plugin --profile desktop add /absolute/path/to/goodandready-dsh-opencode-provider-0.5.21.tgz
+dsh plugin --profile desktop add /absolute/path/to/goodandready-dsh-opencode-provider-0.5.23.tgz
 
 # Or install the published package:
-dsh plugin --profile web add @goodandready/dsh-opencode-provider@0.5.21
+dsh plugin --profile web add @goodandready/dsh-opencode-provider@0.5.23
 ```
 
 Restart dsh (or DSH Desktop), then pick **OpenCode (plugin)** or **OpenCode Go (plugin)** in the model picker.
@@ -147,7 +149,9 @@ Restart dsh (or DSH Desktop), then pick **OpenCode (plugin)** or **OpenCode Go (
 
 ## Verified in DSH Desktop
 
-Tested against DSH **0.2.1-rc.4** (and retained rc.2/rc.3 compatibility):
+Tested against DSH **0.2.2-rc.1** (all 0.2.2 releases are admitted by the compatibility range) and regression-tested on **0.2.1-rc.2/rc.4**: both routes activate, model discovery returns the live catalog, and a real `deepseek-v4.1-flash` request returns `OK`.
+
+The historical Desktop verification below was captured on 0.2.1-rc.4 (and retained rc.2/rc.3 compatibility):
 
 - the plugin loads and `/dsh-opencode/state` returns `200`;
 - the key is picked up from `~/.local/share/opencode/auth.json`;

package/README.ru.md

@@ -93,13 +93,14 @@ graph LR
   - Google `thoughtSignature` в истории вызовов функций.
 - **Пул аккаунтов с интеллектуальной авто-ротацией** до старта генерации:
   - Автоматический переход на следующий аккаунт при 401/403/429 или ответах о превышении квот (400/402/403/429).
-  - Недействительный ключ охлаждается 24 часа, превышение квоты — 10 минут (или по заголовку `Retry-After`), rate limit — 60 секунд (или по `Retry-After`).
+  - До 20 аккаунтов с drag-and-drop приоритетом. Недействительный ключ охлаждается 5 минут, превышение квоты — 10 минут (или по заголовку `Retry-After`), rate limit — 60 секунд (или по `Retry-After`). Аккаунт, обслуживший последний запрос, поднимается первым для следующих диалогов и сохраняется после рестарта.
   - Если запрос уже начал стриминг, он не переключается посреди ответа; следующий запрос берёт следующий готовый ключ.
 - **Sidebar-карточка квот**: отдельный переключатель в разделе **Sidebar** (и в TUI); после сохранения карточка появляется/исчезает сразу, без ожидания 60-секундного polling.
 - **Квоты по каждому аккаунту**: отображение индивидуального расхода за 5 часов, неделю и месяц; активный аккаунт управляет сайдбаром и панелью.
 - **Runway по каждому аккаунту**: оценка времени до исчерпания по среднему темпу каждого окна; отмечается первое окно, которое закончится раньше сброса. Оценка приблизительная: API отдаёт только текущий процент и время сброса.
 - **Онбординг аккаунтов**: **Import all from auth.json** переносит все ключи OpenCode в отдельные credential ref и добавляет аккаунты; **Check all accounts** обновляет квоты и health по каждому аккаунту.
 - **Правила «модель ➔ аккаунт»**: сопоставление по glob-паттернам закрепляет предпочтительный аккаунт за моделью; остальные аккаунты остаются в резерве.
+- **Per-model overrides**: контекст и лимит вывода настраиваются для каждой модели; `0` убирает лимит вывода.
 - **Квотный fallback**: выберите бесплатную модель в настройках; при ошибке квоты Go тот же запрос один раз повторяется через Console/Zen. Если fallback тоже не сработал, возвращается исходная ошибка квоты.
 - **Липкий аккаунт**: последний успешный аккаунт предпочитается в том же диалоге и сохраняется между перезапусками в локальном `sticky.json` (только хешированные session id и fingerprint ключей); явные правила модели, закреплённый аккаунт и квотный fallback имеют приоритет.
 - **Трассировка маршрутизации**: панель квот показывает последние решения (`rule`, `sticky`, `rotation`, `quota-fallback`) без промптов и ключей.
@@ -107,6 +108,7 @@ graph LR
 - **Почасовой Heatmap запросов**: сетка «день недели × час» в локальном времени плюс суточная полоса агрегации. Фильтры переключают отображение между всеми запросами, конкретной моделью и аккаунтом. Лог `~/.dsh/storages/dsh-opencode/requests.jsonl` хранит только метаданные `{at, provider, model, purpose, account, reason, ok, latencyMs, ttftMs, error}` (промпты и ответы не сохраняются).
 - **Стабильный заголовок `x-opencode-session`**: вычисляется из устойчивого ID сессии DSH, обеспечивая одинаковый ключ маршрутизации при повторах и диалоге.
 - **Живой каталог моделей**: автоматическое чтение `/v1/models` с резервным fallback из `models.dev`.
+- **Proxy**: live-параметр `proxyUrl` направляет инференс, каталог и проверку квот через HTTP, HTTPS или SOCKS5; dispatcher переиспользуется до смены адреса.
 - **Оценки лимитов подписки OpenCode Go**: долларовый лимит и расчетное число запросов для окон 5h, недели и месяца для планов Go и Go Plus.
 - **Команда `/opencode`**: быстрый просмотр статуса, квот, аккаунтов, моделей, правил и запуск теста связи `test [model]`.
 - **Настройки в терминале (TUI)**: раздел `/settings → OpenCode` для профиля terminal с ключом, эндпоинтами, allowlist, активным аккаунтом, fallback-моделью, sticky-маршрутизацией и таймаутами.
@@ -129,12 +131,12 @@ npm pack
 Установка из **упакованного tarball**:
 
 ```bash
-dsh plugin --profile web add /absolute/path/to/goodandready-dsh-opencode-provider-0.5.21.tgz
+dsh plugin --profile web add /absolute/path/to/goodandready-dsh-opencode-provider-0.5.23.tgz
 # DSH Desktop:
-dsh plugin --profile desktop add /absolute/path/to/goodandready-dsh-opencode-provider-0.5.21.tgz
+dsh plugin --profile desktop add /absolute/path/to/goodandready-dsh-opencode-provider-0.5.23.tgz
 
 # Либо установка опубликованного пакета из npm:
-dsh plugin --profile web add @goodandready/dsh-opencode-provider@0.5.21
+dsh plugin --profile web add @goodandready/dsh-opencode-provider@0.5.23
 ```
 
 После установки перезапустите dsh (или DSH Desktop) и выберите **OpenCode (plugin)** или **OpenCode Go (plugin)** в списке моделей.

package/README.zh.md

@@ -92,12 +92,13 @@ graph LR
   - Google 历史函数调用中的 `thoughtSignature`。
 - **多账户智能无缝轮换**：
   - 在生成流开始前，遇到 401/403/429 或配额用尽响应（400/402/403/429）自动重试下一个账户。
-  - 无效密钥冷却 24 小时，超额冷却 10 分钟（或遵循 `Retry-After`），速率限制冷却 60 秒。
+  - 最多 20 个账户，支持拖拽调整优先级。无效密钥冷却 5 分钟，超额冷却 10 分钟（或遵循 `Retry-After`），速率限制冷却 60 秒。上次成功服务的账户会提升到首位并在重启后保留。
   - 正在流式传输中的请求不会被打断，下一个请求会自动使用下一个可用密钥。
 - **侧边栏配额卡片**：独立 **Sidebar** 开关（TUI 也可设置）；保存后立即显示/隐藏，无需等待 60 秒轮询。
 - **多账户额度与健康监控**：每行独立展示 5 小时、每周与每月使用情况，活跃账户驱动侧边栏与配额面板。
 - **每账户预计可用时长**：按各窗口的平均消耗速度估算耗尽时间，并标记在重置前最先耗尽的窗口；API 仅提供当前百分比与重置时间，因此为近似值。
 - **模型 ➔ 账户路由规则**：支持 Glob 通配符（如 `*free*`、`deepseek-*`）将模型优先绑定至特定账户，其他账户保留为备用。
+- **按模型覆盖**：可在设置中为每个模型修改上下文窗口和输出上限；`0` 表示该模型省略输出上限。
 - **配额回退**：在设置中选择免费模型；Go 返回配额错误时，同一请求通过 Console/Zen 重试一次。若回退也失败，返回原始配额错误。
 - **会话粘性账户**：同一会话优先使用上次成功的账户，并通过本地 `sticky.json` 跨重启保留（仅哈希后的 session id 和密钥指纹）；显式模型规则、固定账户和配额回退优先。
 - **路由追踪**：配额面板显示最近的账户/模型决策（`rule`、`sticky`、`rotation`、`quota-fallback`），不记录提示词或密钥。
@@ -105,6 +106,7 @@ graph LR
 - **本地时间请求热力图**：星期×小时交互式热力图，排除系统心跳与重复记录，可按模型或账户过滤，仅保存 `{at, provider, model, purpose, account, reason, ok, latencyMs, ttftMs, error}` 元数据，不保存任何提示词与回复。
 - **稳定的 `x-opencode-session` 标头**：根据 DSH 会话 ID 派生，确保重试与连续对话复用相同的网关路由键。
 - **实时模型目录**：自动从 `/v1/models` 同步，并以 `models.dev` 生成文件作为回退。
+- **代理**：一个实时 `proxyUrl` 设置即可让推理、模型目录和额度检查走 HTTP、HTTPS 或 SOCKS5；地址变化前复用 dispatcher。
 - **Go 订阅额度估算**：每月美元额度以及 5 小时、每周和每月的估算请求次数。
 - **`/opencode` 斜杠指令**：快速查询状态、配额、账户、模型及执行联通性测试。
 - **TUI 设置**：`/settings → OpenCode` 支持密钥、端点、allowlist、活动账户、回退模型、粘性路由和超时。
@@ -124,12 +126,12 @@ npm pack
 通过**打包文件 (.tgz)**安装：
 
 ```bash
-dsh plugin --profile web add /absolute/path/to/goodandready-dsh-opencode-provider-0.5.21.tgz
+dsh plugin --profile web add /absolute/path/to/goodandready-dsh-opencode-provider-0.5.23.tgz
 # DSH 桌面端：
-dsh plugin --profile desktop add /absolute/path/to/goodandready-dsh-opencode-provider-0.5.21.tgz
+dsh plugin --profile desktop add /absolute/path/to/goodandready-dsh-opencode-provider-0.5.23.tgz
 
 # 或从 npm 安装已发布版本：
-dsh plugin --profile web add @goodandready/dsh-opencode-provider@0.5.21
+dsh plugin --profile web add @goodandready/dsh-opencode-provider@0.5.23
 ```
 
 安装完成后重启 dsh 或 DSH 桌面端，在模型选择器中选择 **OpenCode (plugin)** 或 **OpenCode Go (plugin)** 即可。

package/lib/client.js

@@ -53,7 +53,7 @@ window.__ModuleLoader__.load({
 				markInvalid: "Invalid key",
 				markTransport: "Transport cooldown",
 				accounts: "Accounts",
-				accountsHint: "The first usable account serves requests; a 401/403/429 or an exhausted quota before the stream starts switches to the next one.",
+				accountsHint: "The first usable account serves requests; a 401/403/429 or an exhausted quota before the stream starts switches to the next one. Drag the handle to change priority.",
 				accountLabel: "Name",
 				accountRef: "Credential ref",
 				accountSource: "Source",
@@ -62,6 +62,8 @@ window.__ModuleLoader__.load({
 				checkAllAccounts: "Check all accounts",
 				importedAll: "Imported all auth.json keys; review the accounts and save.",
 				accountsImported: "Imported keys",
+				dragAccount: "Drag to reorder accounts",
+				accountLimit: "At most 20 accounts are allowed.",
 				removeAccount: "Remove",
 				pinAccount: "Pin",
 				activeAccount: "Pinned account",
@@ -172,7 +174,11 @@ window.__ModuleLoader__.load({
 				usageUrl: "Usage endpoint",
 				requestTimeoutMs: "Header timeout (ms)",
 				streamIdleTimeoutMs: "Stream idle timeout (ms)",
+				proxyUrl: "Proxy URL",
+				proxyUrlHint: "HTTP, HTTPS or SOCKS5 proxy for inference, catalog and usage. Empty disables it. Credentials are not supported.",
 				maxOutputTokens: "Output limit (empty = model default, 0 = omit)",
+				contextOverrideShort: "Context",
+				maxOutputOverrideShort: "Max output",
 				contextWindow: "Context override",
 				providerMigrationNotice: "Provider ids changed in 0.5.7: use OpenCode (plugin) / OpenCode Go (plugin). Sessions recorded with opencode/opencode-go stay on the built-in pi-ai routes; legacy traffic is excluded from plugin analytics.",
 				sidebar: "Sidebar",
@@ -184,6 +190,7 @@ window.__ModuleLoader__.load({
 				routingRule: "Model rule",
 				routingSticky: "Sticky session",
 				routingPinned: "Pinned account",
+				routingPromoted: "Last successful",
 				routingDefault: "Automatic",
 				routingRotation: "Rotation",
 				routingFallback: "Quota fallback",
@@ -259,7 +266,7 @@ window.__ModuleLoader__.load({
 				markInvalid: "密钥无效",
 				markTransport: "传输冷却",
 				accounts: "账户",
-				accountsHint: "第一个可用账户处理请求；流开始前遇到 401/403/429 或配额耗尽时切换到下一个账户。",
+				accountsHint: "第一个可用账户处理请求；流开始前遇到 401/403/429 或配额耗尽时切换到下一个账户。拖动左侧手柄可调整优先级。",
 				accountLabel: "名称",
 				accountRef: "凭据引用",
 				accountSource: "来源",
@@ -268,6 +275,8 @@ window.__ModuleLoader__.load({
 				checkAllAccounts: "检查全部账户",
 				importedAll: "已导入 auth.json 全部密钥；请检查账户并保存。",
 				accountsImported: "已导入密钥",
+				dragAccount: "拖动以调整账户顺序",
+				accountLimit: "最多允许 20 个账户。",
 				removeAccount: "移除",
 				pinAccount: "固定",
 				activeAccount: "固定账户",
@@ -378,7 +387,11 @@ window.__ModuleLoader__.load({
 				usageUrl: "用量端点",
 				requestTimeoutMs: "请求头超时（毫秒）",
 				streamIdleTimeoutMs: "流空闲超时（毫秒）",
+				proxyUrl: "代理地址",
+				proxyUrlHint: "用于推理、目录和用量的 HTTP、HTTPS 或 SOCKS5 代理。留空则禁用。不支持代理认证。",
 				maxOutputTokens: "输出上限（空 = 模型默认，0 = 省略）",
+				contextOverrideShort: "上下文",
+				maxOutputOverrideShort: "最大输出",
 				contextWindow: "上下文覆盖",
 				providerMigrationNotice: "0.5.7 起 provider id 已更改：请选择 OpenCode (plugin) / OpenCode Go (plugin)。使用 opencode/opencode-go 记录的历史会话仍使用内置 pi-ai 路由，旧流量不计入插件分析。",
 				sidebar: "侧边栏",
@@ -390,6 +403,7 @@ window.__ModuleLoader__.load({
 				routingRule: "模型规则",
 				routingSticky: "会话粘性",
 				routingPinned: "固定账户",
+				routingPromoted: "上次成功",
 				routingDefault: "自动",
 				routingRotation: "轮换",
 				routingFallback: "配额回退",
@@ -1160,6 +1174,7 @@ window.__ModuleLoader__.load({
 			if (reason === "rule") return t("routingRule");
 			if (reason === "sticky") return t("routingSticky");
 			if (reason === "pinned") return t("routingPinned");
+			if (reason === "promoted") return t("routingPromoted");
 			if (reason === "rotation") return t("routingRotation");
 			if (reason === "quota-fallback") return t("routingFallback");
 			return t("routingDefault");
@@ -1628,8 +1643,11 @@ window.__ModuleLoader__.load({
 				openCodeAuthEntry: state.config.openCodeAuthEntry,
 				requestTimeoutMs: String(state.config.requestTimeoutMs),
 				streamIdleTimeoutMs: String(state.config.streamIdleTimeoutMs),
+				proxyUrl: state.config.proxyUrl,
 				maxOutputTokens: state.config.maxOutputTokens === null ? "" : String(state.config.maxOutputTokens),
 				contextWindow: state.config.contextWindow === null ? "" : String(state.config.contextWindow),
+				contextWindowOverrides: Object.fromEntries(Object.entries(state.config.contextWindowOverrides).map(([id, value]) => [id, String(value)])),
+				maxOutputTokensOverrides: Object.fromEntries(Object.entries(state.config.maxOutputTokensOverrides).map(([id, value]) => [id, String(value)])),
 				showSidebarQuota: state.config.showSidebarQuota,
 				activeAccount: state.config.activeAccount,
 				accounts: state.config.accounts.map((account) => ({ ...account })),
@@ -1649,6 +1667,17 @@ window.__ModuleLoader__.load({
 			if (!Number.isInteger(value) || value < 0) throw new Error(`Invalid number: ${text}`);
 			return value;
 		}
+		function numberRecordOrNull(record) {
+			const out = {};
+			for (const [id, text] of Object.entries(record)) {
+				const trimmed = text.trim();
+				if (trimmed === "") continue;
+				const value = Number(trimmed);
+				if (!Number.isInteger(value) || value < 0) throw new Error(`Invalid number: ${text}`);
+				out[id] = value;
+			}
+			return out;
+		}
 		function patchOf(draft) {
 			return {
 				apiKeyEnv: draft.apiKeyEnv.trim(),
@@ -1659,8 +1688,11 @@ window.__ModuleLoader__.load({
 				openCodeAuthEntry: draft.openCodeAuthEntry.trim(),
 				requestTimeoutMs: numberOrNull(draft.requestTimeoutMs) ?? 3e5,
 				streamIdleTimeoutMs: numberOrNull(draft.streamIdleTimeoutMs) ?? 3e5,
+				proxyUrl: draft.proxyUrl.trim(),
 				maxOutputTokens: numberOrNull(draft.maxOutputTokens),
 				contextWindow: numberOrNull(draft.contextWindow),
+				contextWindowOverrides: numberRecordOrNull(draft.contextWindowOverrides),
+				maxOutputTokensOverrides: numberRecordOrNull(draft.maxOutputTokensOverrides),
 				showSidebarQuota: draft.showSidebarQuota,
 				activeAccount: draft.activeAccount.trim(),
 				accounts: draft.accounts.map((account) => ({
@@ -1902,6 +1934,7 @@ window.__ModuleLoader__.load({
 			const [testResult, setTestResult] = (0, react.useState)(null);
 			const [importing, setImporting] = (0, react.useState)(false);
 			const [checking, setChecking] = (0, react.useState)(null);
+			const [dragFrom, setDragFrom] = (0, react.useState)(null);
 			const load = (0, react.useCallback)(async () => {
 				setLoading(true);
 				try {
@@ -2039,100 +2072,136 @@ window.__ModuleLoader__.load({
 				const hasPrice = model.cost.input > 0 || model.cost.output > 0;
 				const zen = !model.free && model.limits === null;
 				return /* @__PURE__ */ (0, react_jsx_runtime.jsxs)("div", {
-					className: "oc-model",
-					children: [
-						/* @__PURE__ */ (0, react_jsx_runtime.jsx)("input", {
-							type: "checkbox",
-							checked,
-							onChange: () => toggleModel(model.id)
-						}),
-						/* @__PURE__ */ (0, react_jsx_runtime.jsxs)("span", {
-							className: "oc-model-name",
-							title: model.name + " · " + model.description,
-							children: [
-								model.name,
-								model.free ? /* @__PURE__ */ (0, react_jsx_runtime.jsxs)("span", {
-									className: "oc-badge",
-									title: t("legendFree"),
-									children: [" ", t("free")]
-								}) : null,
-								!model.free && hasPrice ? /* @__PURE__ */ (0, react_jsx_runtime.jsxs)("span", {
-									className: "oc-badge",
-									title: t("legendPrice"),
-									children: [
-										" ",
-										"$",
-										model.cost.input,
-										"/",
-										"$",
-										model.cost.output
-									]
-								}) : null,
-								!model.free && !hasPrice ? /* @__PURE__ */ (0, react_jsx_runtime.jsx)("span", {
-									className: "oc-badge",
-									title: t("legendUnknownPrice"),
-									children: t("priceUnknown")
-								}) : null,
-								model.limits === null ? null : /* @__PURE__ */ (0, react_jsx_runtime.jsxs)("span", {
-									className: "oc-badge",
-									title: describeLimitsFull(model.limits),
-									children: [
-										"Go ",
-										formatRequests(model.limits.go.fiveHour),
-										"/5h"
-									]
-								}),
-								zen ? /* @__PURE__ */ (0, react_jsx_runtime.jsx)("span", {
-									className: "oc-badge",
-									title: t("legendZen"),
-									children: t("zenBadge")
-								}) : null,
-								model.input.includes("image") ? /* @__PURE__ */ (0, react_jsx_runtime.jsxs)("span", {
-									className: "oc-badge",
-									title: t("legendImage"),
-									children: [" ", t("vision")]
-								}) : null
-							]
-						}),
-						/* @__PURE__ */ (0, react_jsx_runtime.jsxs)("select", {
-							className: "oc-select",
-							value: draft.modelTransports[model.id] ?? "",
-							onChange: (event) => {
-								const value = event.target.value;
-								const modelTransports = { ...draft.modelTransports };
-								if (value === "") delete modelTransports[model.id];
-								else modelTransports[model.id] = value;
-								update({ modelTransports });
-							},
-							children: [
-								/* @__PURE__ */ (0, react_jsx_runtime.jsxs)("option", {
-									value: "",
-									children: [
-										t("transportAuto"),
-										" (",
-										model.transport,
-										")"
-									]
-								}),
-								/* @__PURE__ */ (0, react_jsx_runtime.jsx)("option", {
-									value: "openai-chat",
-									children: "openai-chat"
-								}),
-								/* @__PURE__ */ (0, react_jsx_runtime.jsx)("option", {
-									value: "openai-responses",
-									children: "openai-responses"
-								}),
-								/* @__PURE__ */ (0, react_jsx_runtime.jsx)("option", {
-									value: "anthropic",
-									children: "anthropic"
-								}),
-								/* @__PURE__ */ (0, react_jsx_runtime.jsx)("option", {
-									value: "google",
-									children: "google"
-								})
-							]
-						})
-					]
+					className: "oc-model-entry",
+					children: [/* @__PURE__ */ (0, react_jsx_runtime.jsxs)("div", {
+						className: "oc-model",
+						children: [
+							/* @__PURE__ */ (0, react_jsx_runtime.jsx)("input", {
+								type: "checkbox",
+								checked,
+								onChange: () => toggleModel(model.id)
+							}),
+							/* @__PURE__ */ (0, react_jsx_runtime.jsxs)("span", {
+								className: "oc-model-name",
+								title: model.name + " · " + model.description,
+								children: [
+									model.name,
+									model.free ? /* @__PURE__ */ (0, react_jsx_runtime.jsxs)("span", {
+										className: "oc-badge",
+										title: t("legendFree"),
+										children: [" ", t("free")]
+									}) : null,
+									!model.free && hasPrice ? /* @__PURE__ */ (0, react_jsx_runtime.jsxs)("span", {
+										className: "oc-badge",
+										title: t("legendPrice"),
+										children: [
+											" ",
+											"$",
+											model.cost.input,
+											"/",
+											"$",
+											model.cost.output
+										]
+									}) : null,
+									!model.free && !hasPrice ? /* @__PURE__ */ (0, react_jsx_runtime.jsx)("span", {
+										className: "oc-badge",
+										title: t("legendUnknownPrice"),
+										children: t("priceUnknown")
+									}) : null,
+									model.limits === null ? null : /* @__PURE__ */ (0, react_jsx_runtime.jsxs)("span", {
+										className: "oc-badge",
+										title: describeLimitsFull(model.limits),
+										children: [
+											"Go ",
+											formatRequests(model.limits.go.fiveHour),
+											"/5h"
+										]
+									}),
+									zen ? /* @__PURE__ */ (0, react_jsx_runtime.jsx)("span", {
+										className: "oc-badge",
+										title: t("legendZen"),
+										children: t("zenBadge")
+									}) : null,
+									model.input.includes("image") ? /* @__PURE__ */ (0, react_jsx_runtime.jsxs)("span", {
+										className: "oc-badge",
+										title: t("legendImage"),
+										children: [" ", t("vision")]
+									}) : null
+								]
+							}),
+							/* @__PURE__ */ (0, react_jsx_runtime.jsxs)("select", {
+								className: "oc-select",
+								value: draft.modelTransports[model.id] ?? "",
+								onChange: (event) => {
+									const value = event.target.value;
+									const modelTransports = { ...draft.modelTransports };
+									if (value === "") delete modelTransports[model.id];
+									else modelTransports[model.id] = value;
+									update({ modelTransports });
+								},
+								children: [
+									/* @__PURE__ */ (0, react_jsx_runtime.jsxs)("option", {
+										value: "",
+										children: [
+											t("transportAuto"),
+											" (",
+											model.transport,
+											")"
+										]
+									}),
+									/* @__PURE__ */ (0, react_jsx_runtime.jsx)("option", {
+										value: "openai-chat",
+										children: "openai-chat"
+									}),
+									/* @__PURE__ */ (0, react_jsx_runtime.jsx)("option", {
+										value: "openai-responses",
+										children: "openai-responses"
+									}),
+									/* @__PURE__ */ (0, react_jsx_runtime.jsx)("option", {
+										value: "anthropic",
+										children: "anthropic"
+									}),
+									/* @__PURE__ */ (0, react_jsx_runtime.jsx)("option", {
+										value: "google",
+										children: "google"
+									})
+								]
+							})
+						]
+					}), /* @__PURE__ */ (0, react_jsx_runtime.jsxs)("div", {
+						className: "oc-model-overrides",
+						children: [/* @__PURE__ */ (0, react_jsx_runtime.jsxs)("label", {
+							className: "oc-model-override",
+							children: [/* @__PURE__ */ (0, react_jsx_runtime.jsx)("span", { children: t("contextOverrideShort") }), /* @__PURE__ */ (0, react_jsx_runtime.jsx)("input", {
+								className: "oc-input",
+								type: "number",
+								min: 1,
+								placeholder: String(model.contextWindow),
+								value: draft.contextWindowOverrides[model.id] ?? "",
+								onChange: (event) => {
+									const next = { ...draft.contextWindowOverrides };
+									if (event.target.value === "") delete next[model.id];
+									else next[model.id] = event.target.value;
+									update({ contextWindowOverrides: next });
+								}
+							})]
+						}), /* @__PURE__ */ (0, react_jsx_runtime.jsxs)("label", {
+							className: "oc-model-override",
+							children: [/* @__PURE__ */ (0, react_jsx_runtime.jsx)("span", { children: t("maxOutputOverrideShort") }), /* @__PURE__ */ (0, react_jsx_runtime.jsx)("input", {
+								className: "oc-input",
+								type: "number",
+								min: 0,
+								placeholder: String(model.maxTokens),
+								value: draft.maxOutputTokensOverrides[model.id] ?? "",
+								onChange: (event) => {
+									const next = { ...draft.maxOutputTokensOverrides };
+									if (event.target.value === "") delete next[model.id];
+									else next[model.id] = event.target.value;
+									update({ maxOutputTokensOverrides: next });
+								}
+							})]
+						})]
+					})]
 				}, model.id);
 			};
 			const renderSection = (title, hint, rows) => {
@@ -2462,10 +2531,32 @@ window.__ModuleLoader__.load({
 								const draftKey = keyDrafts[index] ?? "";
 								return /* @__PURE__ */ (0, react_jsx_runtime.jsxs)("div", {
 									className: "oc-account-block",
+									onDragOver: (event) => {
+										if (dragFrom !== null) event.preventDefault();
+									},
+									onDrop: () => {
+										if (dragFrom === null || dragFrom === index) return;
+										const accounts = [...draft.accounts];
+										const [moved] = accounts.splice(dragFrom, 1);
+										if (moved !== void 0) accounts.splice(index, 0, moved);
+										update({ accounts });
+										setDragFrom(null);
+									},
 									children: [
 										/* @__PURE__ */ (0, react_jsx_runtime.jsxs)("div", {
 											className: "oc-account",
 											children: [
+												/* @__PURE__ */ (0, react_jsx_runtime.jsx)("span", {
+													className: "oc-drag",
+													draggable: true,
+													title: t("dragAccount"),
+													onDragStart: (event) => {
+														setDragFrom(index);
+														event.dataTransfer.effectAllowed = "move";
+													},
+													onDragEnd: () => setDragFrom(null),
+													children: "⋮⋮"
+												}),
 												/* @__PURE__ */ (0, react_jsx_runtime.jsx)("input", {
 													className: "oc-input",
 													placeholder: t("accountLabel"),
@@ -2586,12 +2677,17 @@ window.__ModuleLoader__.load({
 									/* @__PURE__ */ (0, react_jsx_runtime.jsx)("button", {
 										type: "button",
 										className: "oc-button",
+										disabled: draft.accounts.length >= 20,
 										onClick: () => update({ accounts: [...draft.accounts, {
 											label: "",
 											apiKeyEnv: ""
 										}] }),
 										children: t("addAccount")
 									}),
+									draft.accounts.length >= 20 ? /* @__PURE__ */ (0, react_jsx_runtime.jsx)("span", {
+										className: "oc-hint",
+										children: t("accountLimit")
+									}) : null,
 									/* @__PURE__ */ (0, react_jsx_runtime.jsx)("button", {
 										type: "button",
 										className: "oc-button",
@@ -2944,6 +3040,12 @@ window.__ModuleLoader__.load({
 									value: draft.streamIdleTimeoutMs,
 									onChange: (value) => update({ streamIdleTimeoutMs: value })
 								}),
+								/* @__PURE__ */ (0, react_jsx_runtime.jsx)(Field, {
+									label: t("proxyUrl"),
+									hint: t("proxyUrlHint"),
+									value: draft.proxyUrl,
+									onChange: (value) => update({ proxyUrl: value })
+								}),
 								/* @__PURE__ */ (0, react_jsx_runtime.jsx)(Field, {
 									label: t("maxOutputTokens"),
 									type: "number",
@@ -3092,6 +3194,10 @@ window.__ModuleLoader__.load({
 .oc-bar-fill-bad { background: var(--oc-danger); opacity: 1; }
 .oc-models { display: flex; flex-direction: column; gap: 6px; max-height: 340px; overflow: auto; padding-right: 4px; }
 .oc-model { display: grid; grid-template-columns: 22px minmax(0,1fr) 120px; gap: 8px; align-items: center; font-size: 12px; padding: 3px 0; }
+.oc-model-entry { display: flex; flex-direction: column; gap: 2px; }
+.oc-model-overrides { display: flex; gap: 8px; padding-left: 30px; }
+.oc-model-override { display: flex; align-items: center; gap: 5px; font-size: 10px; color: var(--oc-text-tertiary); }
+.oc-model-override .oc-input { width: 92px; padding: 3px 6px; font-size: 11px; }
 .oc-model-name { min-width: 0; overflow: hidden; text-overflow: ellipsis; white-space: nowrap; }
 .oc-model-legend { display: flex; flex-wrap: wrap; gap: 6px 12px; align-items: center; color: var(--oc-text-secondary); font-size: 11px; }
 .oc-model-legend > span { display: inline-flex; align-items: center; gap: 5px; }
@@ -3100,8 +3206,10 @@ window.__ModuleLoader__.load({
 .oc-model-section-head { display: flex; align-items: center; gap: 8px; padding: 4px 0; font-size: 12px; font-weight: 650; }
 .oc-model-section-head .oc-badge { padding: 0 6px; font-size: 10px; }
 .oc-model-section .oc-hint { margin: 0 0 4px; }
-.oc-account { display: grid; grid-template-columns: minmax(0,1fr) minmax(0,1fr) auto; gap: 8px; align-items: center; }
+.oc-account { display: grid; grid-template-columns: 18px minmax(0,1fr) minmax(0,1fr) auto; gap: 8px; align-items: center; }
 .oc-account-block { display: flex; flex-direction: column; gap: 6px; padding: 9px; border: 1px solid var(--oc-border-soft); border-radius: 8px; background: var(--oc-fill); }
+.oc-drag { cursor: grab; color: var(--oc-text-tertiary); user-select: none; font-size: 13px; line-height: 1; }
+.oc-drag:active { cursor: grabbing; }
 .oc-notice { font-size: 12px; padding: 7px 10px; border-radius: 7px; background: var(--oc-accent-soft); border: 1px solid var(--oc-accent-border); }
 .oc-error { font-size: 12px; padding: 7px 10px; border-radius: 7px; background: var(--oc-danger-soft); border: 1px solid var(--oc-danger-border); color: var(--oc-danger); }
 [class*="_footerActions"]:has(.oc-foot), [class*="_footerActions"]:has(.oc-railButton) { flex-direction: column; }