yuanyuanlove · Oct 11, 2026 · Oct 11, 2026 · Oct 11, 2026 · Oct 11, 2026 · Oct 11, 2026
@@ -251,6 +251,7 @@ Providers can expose a built-in shorthand, such as `agy` for `google-antigravity
 | `responsesPath?` | `string` | Relative resource path for key-auth `openai-responses` requests. It must start with `/` and contain no scheme, query, or fragment. |
 | `chatCompletionsPath?` | `string` | Relative resource path for `openai-chat` requests, the mirror of `responsesPath` and subject to the same shape rules. Needed when one upstream serves Chat Completions and Responses under different prefixes: a per-model wire override changes the adapter and leaves `baseUrl` alone, so without this an opted-in Chat request would be sent to the Responses base. Z.AI is the shipped example. |
 | `allowEncryptedV2AgentTasks?` | `boolean` | Disabled by default. Trust a direct key-auth `openai-responses` provider to consume or relay opaque encrypted V2 sub-agent tasks unchanged. Eligible routes skip `agentTaskRecovery`; all other routes keep the existing recovery or fail-closed behavior. OpenCodex does not decrypt, translate, or recover tasks sent through this opt-in. |
+| `projectResponsesLiteHostedSearchHistory?` | `boolean` | Default off. Set the actual boolean `true` to opt this provider into lossy projection of unreferenced hosted-search history for ordinary requests to exact canonical ChatGPT Codex `/responses` with explicit Responses Lite intent. Omission or `false` retains structural history; strings and numbers are invalid. It does not enable Lite, change transport, or affect native compact, public OpenAI API/other destinations, or existing portable/routed compaction projection. See [Lite hosted-search history](/reference/proxy-formats/#codex-native-ordinary-lite-hosted-search-history). |
 | `upstreamWebsocket?` | `boolean` | Opt-in upstream Responses WebSocket transport for `openai-responses` requests (default false). Honored only for the first-party `https://api.openai.com/v1` upstream; custom-provider endpoints always use bounded HTTP/SSE because Bun cannot enforce an inbound WebSocket message limit before allocating the complete message. For the canonical ChatGPT `openai` provider, omitting it keeps the upstream WebSocket on eligible turns, an explicit `false` sends streaming turns over HTTP/SSE, and provider management rejects `true`; with `false`, native mid-turn steering and injection are unavailable. The field is independent of the client-facing `websockets` setting and changes neither the endpoint nor the credential. Plain HTTP remains on SSE; non-Responses paths and `openai-chat` requests stay on HTTP. |
 | `supportsServiceTier?` | `boolean` | Tri-state canonical Fast capability fallback. `true` publishes Fast in the catalog, satisfies service-tier routing requirements, contributes a supported fingerprint, and lets fast mode inject the provider's canonical wire value on a compatible final adapter. `false` strips the field and never injects, and exact model declarations cannot reopen it. Absent leaves the provider unclassified: fast mode does not inject or normalize a canonical caller value, and caller values obey the final wire's forwarding permission (`chatServiceTier` on Chat; passthrough on Responses). The registry classifies canonical OpenAI (`true`), DeepSeek, and Volcengine Ark (`false`); set it explicitly only for custom gateways that genuinely support tiers. |
 | `responseTierAuthoritative?` | `boolean` | Whether the response tier can confirm or deny Fast. Set `false` explicitly for routes with non-authoritative response metadata; omission preserves existing behavior. This does not enable Fast or change outbound parameters. See [Response service-tier authority](#response-service-tier-authority). |

@@ -32,6 +32,14 @@ the same serialized request and records the recovery in Logs. Other 400 errors,
 custom destinations, cancellations and repeated upload rejections remain failures.
 This does not retry filtered model responses or interrupted streams.
 
+## Codex-native ordinary Lite hosted-search history
+
+Ordinary requests retain structural hosted-search history by default, including Responses Lite requests. To opt in to lossy reference notes, set `providers.openai.projectResponsesLiteHostedSearchHistory: true` in `config.json` (an actual boolean, not the string `"true"`). The per-provider setting defaults to off; omission or `false` keeps structural history. Even when enabled, projection applies only to ordinary requests whose final destination is exact canonical ChatGPT Codex `/responses` and whose effective request intent explicitly selects Responses Lite. It converts unreferenced remaining `web_search_call` history into labeled assistant reference notes. The final `x-openai-internal-codex-responses-lite` header selects Lite only when its value is exactly `true`; an explicit `false` overrides frame metadata. Without a recognized header value, a valid string-map `client_metadata.ws_request_header_x_openai_internal_codex_responses_lite` value of `true` is used, matching the existing WS metadata handling. Non-Lite or unrecognized effective intent retains structural history; malformed metadata does not override an explicit HTTP `true`. Model names, subscription authentication, missing hosted declarations and `additional_tools` alone do not select this policy; no header or metadata is rewritten to make it apply. Once enabled, it applies to new threads/forks, cross-model continuations and cache misses/restarts, without diagnostic settings. Only a valid same-scope bridge memo proves an original function call/result that can be restored first; an ID prefix, model name or missing memo cannot determine the issuer. A hosted cell exactly referenced by a same-input `item_reference.id` is retained. `store:false` removes references earlier, and its remaining hosted history can be projected. Tool declarations/Lite `additional_tools`, choices, client call/result pairs, reasoning/native ciphertext, non-hosted order and caller raw history retain their existing contracts.
+
+Within that opted-in Lite boundary, this is an explicit **lossy history policy**, not a claim that hosted calls are invalid schema or that all upstream errors are fixed. Notes carry untrusted status/action metadata and allowlisted URL/title/type from both action and top-level sources, never instructions or fetched page bodies: 2,048 UTF-16 units per string, 20 entries per list, 65,536 serialized bytes for all projected notes, then an omission note. IDs and unknown fields are dropped from converted cells. HTTP/SSE and WS share the final ordinary builder.
+
+Public OpenAI API and other destinations do not gain ordinary or native-compact projection. Native `/responses/compact` remains its existing endpoint/account-scrub/official opaque-output path and retains structural hosted history. Portable/routed summaries reuse their existing projection trigger and default reference behavior; the shared renderer additionally retains allowlisted top-level sources there. No new retry, source classification or compaction summarizer is introduced.
+
 ## Empty search answers
 
 After hosted search, a clean but empty forced-answer pass receives one additional answer

@@ -92,6 +92,7 @@ selector，而不是分配一个新名称。
 | `requestPacing?` | `{ enabled, requestsPerMinute?, minIntervalMs?, maxConcurrentRequests?, models? }` | 可选的客户端出站请求启动节流，与上游用量、计费和限流指标相互独立。`maxConcurrentRequests` 是限制进行中请求数的正整数，提供商或模型规则都可以只配置此项。提供商限制适用于所有模型，`models` 按上游模型精确 ID 匹配，并可增加延迟或收紧并发限制。排队等待不计入响应头超时。覆盖 HTTP 以及显式适配器 `fetchResponse`/`runTurn` 调用。配置并发上限时，标准 Responses WebSocket 请求改用 HTTP/SSE，以便在响应体完成、出错或取消时释放并发名额。 对包括 Cursor 在内的 `runTurn` 适配器，并发上限计算进行中的回合数，而非物理发送数：同一回合内的 RunSSE 和 BidiAppend 可以重叠，其他回合仍须等待。后续发送仍遵守启动间隔。 |
 | `responsesPath?` | `string` | 用于 key-auth `openai-responses` 请求的相对资源路径。必须以 `/` 开头，且不能包含 scheme、query 或 fragment。 |
 | `chatCompletionsPath?` | `string` | 用于 `openai-chat` 请求的相对资源路径，是 `responsesPath` 的对应项，适用相同的路径规则。当同一上游以不同前缀提供 Chat Completions 和 Responses 时需要此配置：按模型的 wire override 只更换适配器而不改动 `baseUrl`，否则已启用的 Chat 请求会被发送到 Responses base。随附示例为 Z.AI。 |
+| `projectResponsesLiteHostedSearchHistory?` | `boolean` | 默认关闭。仅实际布尔值 `true` 可为此 provider 显式启用有损 hosted-search 历史投影，且只作用于最终目的地精确为标准 ChatGPT Codex `/responses`、明确选择 Responses Lite 的普通请求。省略或设为 `false` 时保留结构化历史；字符串和数字无效。不会启用 Lite、改变传输，也不影响 native compact、公开 OpenAI API／其他目的地或既有 portable/routed compaction 投影。详见[Lite 搜索历史](/zh-cn/reference/proxy-formats/#codex-原生普通-lite-请求的搜索历史)。 |
 | `upstreamWebsocket?` | `boolean` | 为 `openai-responses` 请求选择性启用上游 Responses WebSocket 传输（默认 `false`）。仅对第一方 `https://api.openai.com/v1` 上游生效；自定义提供者端点始终使用有界 HTTP/SSE，因为 Bun 无法在分配完整消息之前对入站 WebSocket 消息实施大小限制。对于规范 ChatGPT `openai` 提供商，省略该字段会在符合条件的轮次使用上游 WebSocket，`false` 通过 HTTP/SSE 发送流式轮次，`true` 会被拒绝；设为 `false` 时，原生轮次中操控与注入不可用。该字段独立于客户端侧的 `websockets` 设置，且不改变端点或凭据。普通 HTTP 仍使用 SSE；非 Responses 路径和 `openai-chat` 请求仍使用 HTTP。 |
 | `supportsServiceTier?` | `boolean` | `service_tier` 能力的三态。`true`：fast 模式可以注入，调用方提供的值也会被保留。`false`：剥离该字段且绝不注入（已明确不支持的上游不会收到它）。未设置：未分类——调用方提供的值原样保留，fast 模式绝不注入。注册表已对官方 OpenAI（`true`）、DeepSeek 和 Volcengine Ark（`false`）分类；仅对真正支持分层的自定义网关显式设置。 |
 | `responseTierAuthoritative?` | `boolean` | 响应等级是否足以确认或否定 Fast。对响应元数据不具权威性的链路显式设为 `false`；未设置时保持原有行为。不会启用 Fast 或修改出站参数。详见[响应服务等级的可信度](#响应服务等级的可信度)。 |

@@ -326,3 +326,11 @@ Anthropic 来源的失败会以 Anthropic 的错误封装呈现，因此该方
 重放的推理项携带的 `encrypted_content` 只有生成它的 provider 和凭据才能读取。如果 opencodex 知道该对话上一次由另一个 provider 处理，它会在发送前移除这个 blob，并保留该项的摘要。如果那个 provider 还使用了不同的 endpoint 或凭据，该项的 `rs_…` id 也会被移除，因为它指向新目标无法查到的项。如果 opencodex 无从得知，例如代理重启之后，新目标会拒绝这个 blob：OpenAI 和 Azure OpenAI 返回 `400 invalid_encrypted_content`。此时 opencodex 会去掉上一个 provider 的推理状态（blob 和 `rs_…` id）后只重发一次请求；保留 id 会导致 `Item with id 'rs_…' not found`。
 
 这项恢复适用于所有使用 Responses 协议的 adapter，因此 `openai-responses` 和 `azure-openai` 的行为相同。恢复成功后，该对话在同一目标上的后续轮次会在接下来五分钟内于首次发送前移除这些状态。重发计入请求的常规发送预算。普通的 400 和 429 不会以这种方式重发，5xx 也不会，只有一个狭窄的例外：对于携带加密工具输出的请求，响应体恰好是该解密拒绝的 502 会得到同样的一次重发。第二次拒绝会原样返回给客户端。遇到这种情况，请在目标 provider 上开始新的对话。
+
+## Codex 原生普通 Lite 请求的搜索历史
+
+普通请求默认保留结构化 hosted-search 历史，Responses Lite 请求也一样。若要选择有损参考说明，请在 `config.json` 中设置 `providers.openai.projectResponsesLiteHostedSearchHistory: true`（必须是布尔值，不能是字符串 `"true"`）。该设置按 provider 配置且默认关闭；省略或设为 `false` 时保留结构化历史。即使已启用，也只有最终目的地精确为标准 ChatGPT Codex `/responses` 且当前有效意图明确选择 Responses Lite 的普通请求，才会把未被同输入引用的剩余 `web_search_call` 历史转为带标签的 assistant 参考说明。最终 `x-openai-internal-codex-responses-lite` header 值精确为 `true` 时选择 Lite；明确的 `false` 优先于 frame metadata。没有可识别 header 值时，采用有效字符串 map 中 `client_metadata.ws_request_header_x_openai_internal_codex_responses_lite` 的 `true` 值，沿用既有 WS metadata 语义。最终有效意图为非 Lite 或无法识别时保留结构化历史；格式无效的 metadata 不会覆盖明确的 HTTP `true`。模型名、订阅鉴权、hosted 声明缺失或仅有 `additional_tools` 都不能选择这项策略；不会为触发转换而改写 header 或 metadata。启用后适用于新 thread、fork、跨模型续聊、缓存未命中或重启，不依赖诊断设置。只有有效同 scope bridge memo 能证明并先恢复原函数调用及结果；ID 前缀、模型名或 memo 缺失不能判断来源。若同输入的 `item_reference.id` 精确指向某 hosted cell，该 cell 保留；`store:false` 会先移除引用，剩余 hosted 历史可投影。工具声明、Lite `additional_tools`、选择、call/result 配对、推理/原生密文、非 hosted 顺序及调用方 raw 历史保持原契约。
+
+这是仅在上述显式启用的 Lite 边界内采用的**有损历史策略**，不是 hosted schema 非法或所有上游错误已修复的断言。说明标为不可信参考数据，复制 status/action 及 action 或顶层 sources 中允许的 URL/title/type，不是指令或网页正文；每字符串最多 2,048 个 UTF-16 单元、每列表 20 项、全部已投影说明共 65,536 字节，超出的 action 用省略说明替代，转换项不复制 ID 或未知字段。HTTP/SSE 与 WS 共用普通请求的最终 builder。
+
+公开 OpenAI API 及其他目的地不新增普通或 native compact 投影。原生 `/responses/compact` 保持既有端点、账户清理及官方 opaque 输出路径，保留结构化 hosted 历史。portable/routed 摘要仍沿用原有触发条件与默认引用行为；共享 renderer 在这些既有路径中也补充顶层 sources 白名单保留。本次不增加重试、来源分类或新的 compaction summarizer。
@@ -72,6 +72,7 @@ ocx models provider openrouter on
 | `requestPacing?` | `{ enabled, requestsPerMinute?, minIntervalMs?, maxConcurrentRequests?, models? }` | 選用的用戶端出站請求啟動節流，與上游用量、計費及限流指標彼此獨立。`maxConcurrentRequests` 是限制進行中請求數的正整數，供應商或模型規則都可單獨設定此項。供應商限制適用於所有模型，`models` 依上游模型精確 ID 比對，並可增加延遲或收緊並發限制。排隊等待不計入回應標頭逾時。涵蓋 HTTP 及明確的適配器 `fetchResponse`/`runTurn` 呼叫。設定並行上限時，標準 Responses WebSocket 請求改用 HTTP/SSE，以便在回應本文完成、出錯或取消時釋放並行名額。 對包含 Cursor 的 `runTurn` 適配器，並行上限計算進行中的回合數，而非實體傳送數：同一回合內的 RunSSE 與 BidiAppend 可重疊，其他回合仍須等待。後續傳送仍遵守啟動間隔。 |
 | `responsesPath?` | `string` | Key-auth `openai-responses` 請求的相對資源路徑。必須以 `/` 開頭且不含 scheme、query 或 fragment。 |
 | `chatCompletionsPath?` | `string` | `openai-chat` 請求的相對資源路徑，為 `responsesPath` 的對應項，適用相同的路徑規則。當同一上游以不同前綴提供 Chat Completions 與 Responses 時需要此設定：按模型的 wire override 只更換適配器而不改動 `baseUrl`，否則已啟用的 Chat 請求會送往 Responses base。隨附範例為 Z.AI。 |
+| `projectResponsesLiteHostedSearchHistory?` | `boolean` | 預設關閉。僅實際布林值 `true` 可為此供應商明確啟用有損 hosted-search 歷史投影，且只作用於最終目的地精確為標準 ChatGPT Codex `/responses`、明確選擇 Responses Lite 的一般請求。省略或設為 `false` 時保留結構化歷史；字串和數字無效。不會啟用 Lite、變更傳輸，也不影響 native compact、公開 OpenAI API／其他目的地或既有 portable/routed compaction 投影。詳見[Lite 搜尋歷史](/zh-tw/reference/proxy-formats/#codex-原生一般-lite-請求的搜尋歷史)。 |
 | `upstreamWebsocket?` | `boolean` | 為 `openai-responses` 請求選用上游 Responses WebSocket 傳輸（預設 `false`）。僅對第一方 `https://api.openai.com/v1` 上游生效；自訂供應商端點一律使用有界 HTTP/SSE，因為 Bun 無法在配置完整訊息之前對傳入 WebSocket 訊息套用大小限制。對於規範 ChatGPT `openai` 供應商，省略此欄位會在符合條件的回合使用上游 WebSocket，`false` 會以 HTTP/SSE 傳送串流回合，`true` 會被拒絕；設為 `false` 時，原生回合中操控與注入無法使用。此欄位獨立於用戶端 `websockets` 設定，且不會變更端點或認證資料。一般 HTTP 仍使用 SSE；非 Responses 路徑與 `openai-chat` 請求仍使用 HTTP。 |
 | `disabled?` | `boolean` | 將供應商保留在磁碟上但排除於路由與模型／目錄清單。 |
 | `apiKey?` | `string` | API 金鑰，或在請求時解析的 `${ENV_VAR}` / `$ENV_VAR` 參考。 |