omnigateway 0.7.1 → 0.7.2

package/README.md

@@ -21,9 +21,10 @@ omni start
 
 ## What it does
 
-- **Speaks both dialects.** `POST /v1/messages` (Anthropic) and
-  `POST /v1/chat/completions` (OpenAI), including streaming, translated to
-  whichever provider actually serves the request.
+- **Speaks three dialects.** `POST /v1/messages` (Anthropic),
+  `POST /v1/chat/completions` (OpenAI) and `POST /v1/responses` (OpenAI
+  Responses, which is the only API Codex CLI speaks), including streaming,
+  translated to whichever provider actually serves the request.
 - **Filters bulky tool history, optionally.** Built-in RTK filters shorten
   eligible large or repetitive shell and recognizable command output before
   provider dispatch, preserve errors and non-tool-result content, and default off.
@@ -289,6 +290,7 @@ Everything above is also available in the browser at
 | --- | --- | --- |
 | `POST` | `/v1/messages` | Anthropic Messages API |
 | `POST` | `/v1/chat/completions` | OpenAI Chat Completions API |
+| `POST` | `/v1/responses` | OpenAI Responses API (stateless: no stored responses, no `previous_response_id`) |
 | `POST` | `/v1/messages/count_tokens` | Anthropic-compatible local token estimation (authenticated) |
 | `GET` | `/v1/models` | Listing in both dialects, filtered by your key's allowlist |
 | `GET` | `/health` | Unauthenticated liveness check |
@@ -315,7 +317,7 @@ Most tools that accept a custom base URL work unchanged: set it to
 
 Every response carries the limit headers of the surface you asked on, so an SDK
 backs off using the code it already ships — the Anthropic dialect on
-`/v1/messages`, the OpenAI one on `/v1/chat/completions`:
+`/v1/messages`, the OpenAI one on `/v1/chat/completions` and `/v1/responses`:
 
 ```http
 anthropic-ratelimit-requests-limit: 2000        x-ratelimit-limit-requests: 2000

package/bin/omni.js

@@ -1753,6 +1753,10 @@ class GatewayError extends Error {
 }
 // packages/ir/src/logger.ts
 var LEVELS = { debug: 10, info: 20, warn: 30, error: 40 };
+function safeToken(value) {
+  const text = typeof value === "string" ? value : String(value);
+  return /^[A-Za-z0-9_.:/+-]{1,64}$/.test(text) ? text : "(unprintable)";
+}
 var ANSI2 = /\u001B\[[0-9;]*m/g;
 var HEAD = /^(\S+) (DEBUG|INFO|WARN|ERROR) {1,2}(.*)$/;
 var FIELDS = / {2}(?=[a-zA-Z]+=)/;
@@ -20642,10 +20646,10 @@ function toCustomResponsesWire(req, model) {
     note2("custom:context-1m-dropped");
   for (const message of req.messages) {
     const parts = [];
-    const inlined = message.role === "system";
-    if (inlined)
-      note2("custom:system-turn-inlined");
-    const role = inlined ? "user" : message.role;
+    const asDeveloper = message.role === "system";
+    if (asDeveloper)
+      note2("custom:system-turn-as-developer");
+    const role = asDeveloper ? "developer" : message.role;
     const flush = () => {
       if (parts.length === 0)
         return;
@@ -20657,9 +20661,7 @@ function toCustomResponsesWire(req, model) {
         case "text":
           parts.push({
             type: role === "assistant" ? "output_text" : "input_text",
-            text: inlined ? `<system-reminder>
-${block.text}
-</system-reminder>` : block.text
+            text: block.text
           });
           break;
         case "image":
@@ -20737,7 +20739,7 @@ function metadata(data, fail) {
 function endpointUrl(origin, basePath, protocol) {
   const suffix = protocol === "chat_completions" ? "chat/completions" : "responses";
   const base = `${origin}${basePath}`.replace(/\/+$/, "");
-  return base.endsWith("/v1") ? `${base}/${suffix}` : `${base}/v1/${suffix}`;
+  return /\/v\d+$/i.test(base) ? `${base}/${suffix}` : `${base}/v1/${suffix}`;
 }
 function protocolOf(state, fail) {
   if (state !== null && typeof state === "object") {
@@ -21006,10 +21008,10 @@ function toGrokWire(req, model) {
     note2("grok:context-1m-dropped");
   for (const message of req.messages) {
     const parts = [];
-    const inlined = message.role === "system";
-    if (inlined)
-      note2("grok:system-turn-inlined");
-    const role = inlined ? "user" : message.role;
+    const asDeveloper = message.role === "system";
+    if (asDeveloper)
+      note2("grok:system-turn-as-developer");
+    const role = asDeveloper ? "developer" : message.role;
     const flush = () => {
       if (parts.length === 0)
         return;
@@ -21021,9 +21023,7 @@ function toGrokWire(req, model) {
         case "text":
           parts.push({
             type: role === "assistant" ? "output_text" : "input_text",
-            text: inlined ? `<system-reminder>
-${block.text}
-</system-reminder>` : block.text
+            text: block.text
           });
           break;
         case "image":
@@ -21815,6 +21815,7 @@ var ERROR_CODE2 = {
   content_policy_violation: "CONTENT_FILTER"
 };
 var KNOWN_EVENTS3 = new Set([
+  "keepalive",
   "response.created",
   "response.queued",
   "response.in_progress",
@@ -21848,7 +21849,8 @@ function json7(data) {
     return null;
   }
 }
-async function* decodeResponses(messages) {
+async function* decodeResponses(messages, options = {}) {
+  const nativeReasoning = options.nativeReasoning === true;
   const indices = new Map;
   let next = 0;
   const irIndex = (outputIndex, contentIndex = 0) => {
@@ -21893,7 +21895,7 @@ async function* decodeResponses(messages) {
           yield {
             type: "blockStart",
             index: irIndex(d.output_index ?? 0),
-            block: { type: "thinking" }
+            block: nativeReasoning ? { type: "providerNative", provider: "openai", blockType: "reasoning", data: {} } : { type: "thinking" }
           };
         } else if (item.type === "function_call") {
           sawToolCall = true;
@@ -21926,7 +21928,12 @@ async function* decodeResponses(messages) {
         yield {
           type: "blockDelta",
           index: irIndex(d.output_index ?? 0),
-          delta: { type: "thinking", text: String(d.delta ?? "") }
+          delta: nativeReasoning ? {
+            type: "providerNative",
+            provider: "openai",
+            deltaType: "reasoning_summary_text.delta",
+            data: { text: String(d.delta ?? "") }
+          } : { type: "thinking", text: String(d.delta ?? "") }
         };
         break;
       case "response.function_call_arguments.delta":
@@ -21942,6 +21949,20 @@ async function* decodeResponses(messages) {
       case "response.output_item.done": {
         const outputIndex = d.output_index ?? 0;
         if (ownsBlock.delete(outputIndex)) {
+          const item = d.item ?? {};
+          if (nativeReasoning && item.type === "reasoning") {
+            yield {
+              type: "blockDelta",
+              index: irIndex(outputIndex),
+              delta: {
+                type: "providerNative",
+                provider: "openai",
+                deltaType: "response.output_item.done",
+                fold: "merge",
+                data: item
+              }
+            };
+          }
           yield { type: "blockEnd", index: irIndex(outputIndex) };
         }
         break;
@@ -22015,6 +22036,23 @@ function cacheKey(req, instructions, firstInput) {
   const stable = req.conversationId !== undefined && req.conversationId.length > 0 ? JSON.stringify({ conversation: req.conversationId }) : JSON.stringify({ instructions, firstInput: firstInput ?? null });
   return createHash3("sha256").update(stable).digest("hex").slice(0, 32);
 }
+function repairOrphanedCalls(input) {
+  const answered = new Set;
+  for (const item of input) {
+    const record2 = item;
+    if (record2.type === "function_call_output" && typeof record2.call_id === "string") {
+      answered.add(record2.call_id);
+    }
+  }
+  for (let i = input.length - 1;i >= 0; i--) {
+    const record2 = input[i];
+    if (record2.type !== "function_call" || typeof record2.call_id !== "string")
+      continue;
+    if (answered.has(record2.call_id))
+      continue;
+    input.splice(i + 1, 0, { type: "function_call_output", call_id: record2.call_id, output: "" });
+  }
+}
 function encodeToolChoice5(c) {
   switch (c.type) {
     case "auto":
@@ -22038,10 +22076,10 @@ function toResponsesWire(req, model, opts = { oauth: false }) {
     note2("openai:context-1m-dropped");
   for (const message of req.messages) {
     const parts = [];
-    const inlined = message.role === "system";
-    if (inlined)
-      note2("openai:system-turn-inlined");
-    const role = inlined ? "user" : message.role;
+    const asDeveloper = message.role === "system";
+    if (asDeveloper)
+      note2("openai:system-turn-as-developer");
+    const role = asDeveloper ? "developer" : message.role;
     const flush = () => {
       if (parts.length === 0)
         return;
@@ -22053,9 +22091,7 @@ function toResponsesWire(req, model, opts = { oauth: false }) {
         case "text":
           parts.push({
             type: role === "assistant" ? "output_text" : "input_text",
-            text: inlined ? `<system-reminder>
-${block.text}
-</system-reminder>` : block.text
+            text: block.text
           });
           break;
         case "image":
@@ -22087,12 +22123,19 @@ ${block.text}
           });
           break;
         case "providerNative":
+          if (block.provider === "openai") {
+            flush();
+            const { id: _id, ...rest } = block.data;
+            input.push({ type: block.blockType, ...rest });
+            break;
+          }
           note2("openai:anthropic-native-block-dropped");
           break;
       }
     }
     flush();
   }
+  repairOrphanedCalls(input);
   const body = { model, input, stream: req.stream, store: false };
   const instructions = systemText(req.system, "openai", note2);
   if (instructions !== undefined && instructions.length > 0)
@@ -22156,7 +22199,10 @@ var openaiCodec = {
       throw input.fail("AUTH", "openai credential has no token");
     }
     const headers = orderHeaders(mergeHeaders(openaiProfile.headers, protocol), openaiProfile.order);
+    const include = body.include;
+    const nativeReasoning = Array.isArray(include) && include.includes("reasoning.encrypted_content");
     return {
+      decodeState: { nativeReasoning },
       request: {
         url: oauth ? OAUTH_URL2 : API_URL3,
         method: "POST",
@@ -22166,8 +22212,9 @@ var openaiCodec = {
       degradations
     };
   },
-  decode({ body }) {
-    return decodeResponses(parseSse(body));
+  decode({ body, decodeState }) {
+    const state = decodeState;
+    return decodeResponses(parseSse(body), { nativeReasoning: state?.nativeReasoning === true });
   }
 };
 
@@ -22344,8 +22391,10 @@ function parseOrThrow(schema, body2) {
   if (result.success)
     return result.data;
   const issue2 = result.error.issues[0];
-  const path = issue2?.path.join(".") ?? "(root)";
-  throw new GatewayError("BAD_REQUEST", `${path}: ${issue2?.message ?? "invalid request"}`);
+  const segments = issue2?.path ?? [];
+  const path = segments.length === 0 ? "(root)" : segments.map((s) => safeToken(String(s))).join(".");
+  const detail = issue2 === undefined ? "invalid request" : issue2.code === "unrecognized_keys" ? `unrecognized key "${safeToken(issue2.keys[0])}"` : issue2.message;
+  throw new GatewayError("BAD_REQUEST", `${path}: ${detail}`);
 }
 function optionalNumber(value, fallback) {
   if (value === undefined)
@@ -27062,7 +27111,7 @@ function createServiceDeps(input) {
 }
 
 // apps/cli/src/version.ts
-var VERSION = "0.7.1";
+var VERSION = "0.7.2";
 
 // apps/cli/src/commands/service.ts
 function gatewayArgv(root) {