fix(models): honor per-model targetFormat for custom models (#2905)

Custom models (added via the UI) on opencode-go / openai-compatible nodes always
routed as OpenAI-compatible because there was no per-model targetFormat: addCustomModel
didn't accept it, the API schema stripped it, and getModelTargetFormat (static-registry
only) never saw it — so a custom model needing the Anthropic Messages shape fell back to
the provider default 'openai'.

Thread an optional targetFormat through addCustomModel / replaceCustomModels /
updateCustomModel + the providerModelMutationSchema + the POST/PUT route, surface it from
getModelInfo (one combined custom-model metadata lookup alongside apiFormat), and use it
in chatCore's targetFormat resolution before the provider default.

Closes #2905

CHANGELOG.md

@@ -59,6 +59,12 @@
 
 ### Fixed
 
+- **models/custom:** custom models can now carry a per-model `targetFormat`
+  override (e.g. an opencode-go custom model that must use the Anthropic Messages
+  shape). Previously custom models always routed as OpenAI-compatible because
+  `targetFormat` was neither persisted nor consulted at routing time. Threaded
+  through `addCustomModel`/`replaceCustomModels`/`updateCustomModel`, the API
+  schema/route, `getModelInfo`, and chatCore's targetFormat resolution. (#2905)
 - **providers/pollinations:** route to `gen.pollinations.ai/v1` instead of the
   retired `text.pollinations.ai` host, which now returns `404 "legacy API"` for
   all models. The gen gateway is the current OpenAI-compatible endpoint. (#2987)

open-sse/handlers/chatCore.ts

@@ -1495,6 +1495,15 @@ export async function handleChatCore({
         ? ((modelInfo as { apiFormat?: string }).apiFormat as string)
         : undefined
       : undefined;
+  // #2905: per-model wire-format override for custom models, injected by
+  // getModelInfo. Custom models are not in the static registry, so
+  // getModelTargetFormat() can't see this — use it before the provider default.
+  const customModelTargetFormat: string | undefined =
+    modelInfo && typeof modelInfo === "object" && "targetFormat" in modelInfo
+      ? typeof (modelInfo as { targetFormat?: unknown }).targetFormat === "string"
+        ? ((modelInfo as { targetFormat?: string }).targetFormat as string)
+        : undefined
+      : undefined;
   const requestedModel =
     typeof body?.model === "string" && body.model.trim().length > 0 ? body.model : model;
   const isModelScope = () => isModelScopeProvider(provider, credentials?.providerSpecificData);
@@ -1876,7 +1885,9 @@ export async function handleChatCore({
   const targetFormat =
     apiFormat === "responses"
       ? FORMATS.OPENAI_RESPONSES
-      : modelTargetFormat || getTargetFormat(provider, credentials?.providerSpecificData);
+      : modelTargetFormat ||
+        customModelTargetFormat ||
+        getTargetFormat(provider, credentials?.providerSpecificData);
 
   const initialProviderRequest =
     body && typeof body === "object" && !Array.isArray(body)

src/app/api/provider-models/route.ts

@@ -96,7 +96,8 @@ export async function POST(request) {
     if (isValidationFailure(validation)) {
       return Response.json({ error: validation.error }, { status: 400 });
     }
-    const { provider, modelId, modelName, source, apiFormat, supportedEndpoints } = validation.data;
+    const { provider, modelId, modelName, source, apiFormat, supportedEndpoints, targetFormat } =
+      validation.data;
 
     const model = await addCustomModel(
       provider,
@@ -104,7 +105,8 @@ export async function POST(request) {
       modelName,
       source || "manual",
       apiFormat,
-      supportedEndpoints
+      supportedEndpoints,
+      targetFormat
     );
     return Response.json({ model });
   } catch (error) {
@@ -150,6 +152,7 @@ export async function PUT(request) {
       modelName,
       apiFormat,
       supportedEndpoints,
+      targetFormat,
       normalizeToolCallId,
       preserveOpenAIDeveloperRole,
       upstreamHeaders,
@@ -161,6 +164,7 @@ export async function PUT(request) {
     if ("modelName" in raw) updates.modelName = modelName;
     if ("apiFormat" in raw) updates.apiFormat = apiFormat;
     if ("supportedEndpoints" in raw) updates.supportedEndpoints = supportedEndpoints;
+    if ("targetFormat" in raw) updates.targetFormat = targetFormat;
     if ("normalizeToolCallId" in raw) updates.normalizeToolCallId = normalizeToolCallId;
     if ("preserveOpenAIDeveloperRole" in raw)
       updates.preserveOpenAIDeveloperRole = preserveOpenAIDeveloperRole;

src/lib/db/models.ts

@@ -355,7 +355,11 @@ export async function addCustomModel(
     | "audio-transcriptions"
     | "audio-speech"
     | "images-generations" = "chat-completions",
-  supportedEndpoints: string[] = ["chat"]
+  supportedEndpoints: string[] = ["chat"],
+  // #2905: optional per-model wire format override (e.g. "claude" for an
+  // opencode-go custom model). When unset, routing falls back to the provider
+  // default format.
+  targetFormat?: string
 ) {
   const db = getDbInstance();
   const row = db
@@ -373,6 +377,7 @@ export async function addCustomModel(
     source,
     apiFormat,
     supportedEndpoints,
+    ...(targetFormat ? { targetFormat } : {}),
   };
   models.push(model);
   db.prepare(
@@ -398,6 +403,7 @@ export async function replaceCustomModels(
     outputTokenLimit?: number;
     description?: string;
     supportsThinking?: boolean;
+    targetFormat?: string;
   }>,
   { allowEmpty = false }: { allowEmpty?: boolean } = {}
 ) {
@@ -427,6 +433,12 @@ export async function replaceCustomModels(
       source: m.source || "auto-sync",
       apiFormat: m.apiFormat || (prev as any)?.apiFormat || "chat-completions",
       supportedEndpoints: m.supportedEndpoints || (prev as any)?.supportedEndpoints || ["chat"],
+      // #2905: preserve a per-model targetFormat override (new value wins, else prev).
+      ...(m.targetFormat
+        ? { targetFormat: m.targetFormat }
+        : (prev as any)?.targetFormat
+          ? { targetFormat: (prev as any).targetFormat }
+          : {}),
       // Preserve metadata from provider API (or previous sync)
       ...(m.inputTokenLimit != null
         ? { inputTokenLimit: m.inputTokenLimit }
@@ -815,6 +827,7 @@ export async function updateCustomModel(
     ...current,
     ...(updates.modelName !== undefined ? { name: updates.modelName || current.name } : {}),
     ...(updates.apiFormat !== undefined ? { apiFormat: updates.apiFormat } : {}),
+    ...(updates.targetFormat !== undefined ? { targetFormat: updates.targetFormat } : {}),
     ...(updates.supportedEndpoints !== undefined
       ? { supportedEndpoints: updates.supportedEndpoints }
       : {}),

src/shared/validation/schemas.ts

@@ -1048,6 +1048,11 @@ export const providerModelMutationSchema = z.object({
       ])
     )
     .default(["chat"]),
+  // #2905: optional per-model wire format override for custom models (e.g. a
+  // custom opencode-go model that must use the Anthropic Messages shape).
+  targetFormat: z
+    .enum(["openai", "openai-responses", "claude", "gemini", "gemini-cli", "antigravity"])
+    .optional(),
   normalizeToolCallId: z.boolean().optional(),
   preserveOpenAIDeveloperRole: z.boolean().nullable().optional(),
   upstreamHeaders: upstreamHeadersRecordSchema.nullable().optional(),

src/sse/services/model.ts

@@ -48,20 +48,25 @@ export async function resolveModelAlias(alias) {
 }
 
 /**
- * Look up the apiFormat for a custom model from the DB.
- * Returns "responses" if the model is configured for the Responses API, otherwise undefined.
+ * Look up custom-model metadata from the DB in a single read:
+ *  - apiFormat: "responses" when the model is configured for the Responses API.
+ *  - targetFormat: the optional per-model wire format override (#2905).
  */
-async function lookupCustomModelApiFormat(
+async function lookupCustomModelMeta(
   providerId: string,
   modelId: string
-): Promise<string | undefined> {
+): Promise<{ apiFormat?: string; targetFormat?: string }> {
   try {
     const models = await getCustomModels(providerId);
-    if (!Array.isArray(models)) return undefined;
+    if (!Array.isArray(models)) return {};
     const match = models.find((m: any) => m.id === modelId);
-    return match?.apiFormat === "responses" ? "responses" : undefined;
+    if (!match) return {};
+    return {
+      apiFormat: match.apiFormat === "responses" ? "responses" : undefined,
+      targetFormat: typeof match.targetFormat === "string" ? match.targetFormat : undefined,
+    };
   } catch {
-    return undefined;
+    return {};
   }
 }
 
@@ -74,11 +79,15 @@ export async function getModelInfo(modelStr) {
 
   const attachCustomApiFormat = async (info: any) => {
     if (!info?.provider || !info?.model) return info;
-    const apiFormat = await lookupCustomModelApiFormat(String(info.provider), String(info.model));
-    if (apiFormat) {
+    const { apiFormat, targetFormat } = await lookupCustomModelMeta(
+      String(info.provider),
+      String(info.model)
+    );
+    if (apiFormat || targetFormat) {
       return {
         ...info,
-        apiFormat,
+        ...(apiFormat && { apiFormat }),
+        ...(targetFormat && { targetFormat }),
       };
     }
     return info;
@@ -98,7 +107,7 @@ export async function getModelInfo(modelStr) {
       (node) => node.prefix === prefixToCheck || node.id === prefixToCheck
     );
     if (matchedOpenAI) {
-      const apiFormat = await lookupCustomModelApiFormat(
+      const { apiFormat, targetFormat } = await lookupCustomModelMeta(
         matchedOpenAI.id as string,
         parsed.model as string
       );
@@ -107,6 +116,7 @@ export async function getModelInfo(modelStr) {
         model: parsed.model,
         extendedContext,
         ...(apiFormat && { apiFormat }),
+        ...(targetFormat && { targetFormat }),
       };
     }
 
@@ -116,7 +126,7 @@ export async function getModelInfo(modelStr) {
       (node) => node.prefix === prefixToCheck || node.id === prefixToCheck
     );
     if (matchedAnthropic) {
-      const apiFormat = await lookupCustomModelApiFormat(
+      const { apiFormat, targetFormat } = await lookupCustomModelMeta(
         matchedAnthropic.id as string,
         parsed.model as string
       );
@@ -125,6 +135,7 @@ export async function getModelInfo(modelStr) {
         model: parsed.model,
         extendedContext,
         ...(apiFormat && { apiFormat }),
+        ...(targetFormat && { targetFormat }),
       };
     }
 

tests/unit/custom-model-target-format.test.ts

@@ -0,0 +1,83 @@
+/**
+ * Issue #2905 — custom models (added via the UI) on opencode-go / openai-compatible
+ * nodes always routed as OpenAI-compatible because there was no way to set a
+ * per-model `targetFormat`. `addCustomModel` didn't accept it, the API schema
+ * stripped it, and routing (`getModelTargetFormat`, static-registry-only) never
+ * saw it — so a custom model that needs the Anthropic Messages shape fell back
+ * to the provider default ("openai").
+ *
+ * This test verifies the persistence + getModelInfo-injection chain: a custom
+ * model saved with targetFormat: "claude" is surfaced on the resolved modelInfo
+ * (which chatCore then uses before the provider default).
+ */
+import test from "node:test";
+import assert from "node:assert/strict";
+import fs from "node:fs";
+import os from "node:os";
+import path from "node:path";
+
+const TEST_DATA_DIR = fs.mkdtempSync(path.join(os.tmpdir(), "omniroute-custom-target-format-"));
+process.env.DATA_DIR = TEST_DATA_DIR;
+
+const core = await import("../../src/lib/db/core.ts");
+const providersDb = await import("../../src/lib/db/providers.ts");
+const modelsDb = await import("../../src/lib/db/models.ts");
+const { getModelInfo } = await import("../../src/sse/services/model.ts");
+
+test.before(async () => {
+  await providersDb.createProviderNode({
+    id: "openai-compatible-2905",
+    type: "openai-compatible",
+    name: "Gateway 2905",
+    prefix: "g29",
+    baseUrl: "https://proxy.example.com",
+    chatPath: "/v1/chat/completions",
+    modelsPath: "/v1/models",
+  });
+});
+
+test.after(() => {
+  core.resetDbInstance();
+  fs.rmSync(TEST_DATA_DIR, { recursive: true, force: true });
+});
+
+test("#2905 addCustomModel persists targetFormat", async () => {
+  await modelsDb.addCustomModel(
+    "openai-compatible-2905",
+    "my-claude-model",
+    "My Claude Model",
+    "manual",
+    "chat-completions",
+    ["chat"],
+    "claude"
+  );
+  const models = (await modelsDb.getCustomModels("openai-compatible-2905")) as Array<{
+    id: string;
+    targetFormat?: string;
+  }>;
+  const saved = models.find((m) => m.id === "my-claude-model");
+  assert.ok(saved, "custom model must be saved");
+  assert.equal(saved.targetFormat, "claude", "targetFormat must be persisted");
+});
+
+test("#2905 getModelInfo surfaces the custom model targetFormat", async () => {
+  const info = (await getModelInfo("g29/my-claude-model")) as {
+    provider?: string;
+    targetFormat?: string;
+  };
+  assert.equal(info.provider, "openai-compatible-2905", "must resolve to the custom node");
+  assert.equal(info.targetFormat, "claude", "getModelInfo must inject the custom targetFormat");
+});
+
+test("#2905 a custom model without targetFormat surfaces none (provider default applies)", async () => {
+  await modelsDb.addCustomModel(
+    "openai-compatible-2905",
+    "plain-model",
+    "Plain Model",
+    "manual",
+    "chat-completions",
+    ["chat"]
+  );
+  const info = (await getModelInfo("g29/plain-model")) as { targetFormat?: string };
+  assert.equal(info.targetFormat, undefined, "no targetFormat → falls back to provider default");
+});