pjandro/OmniRoute
1
0
Fork 0
mirror of https://github.com/diegosouzapw/OmniRoute.git synced 2026-07-31 04:12:10 +03:00
Code Issues Packages Projects Releases Wiki Activity

fix(routing): codex bare-name disambiguation + family-native fallback (#1933)

Browse Source 
Integrated into release/v3.7.9
This commit is contained in:
smartenok-ops
2026-05-04 08:40:49 -04:00
committed by GitHub
parent 9577a8d9e8
commit ec0ce5bed7
1 changed files with 76 additions and 11 deletions
Download Patch File Download Diff File Expand all files Collapse all files

87
src/sse/handlers/chatHelpers.ts
View File

@@ -29,22 +29,87 @@ import { logProxyEvent } from "../../lib/proxyLogger";
import { logTranslationEvent } from "../../lib/translatorEvents";
import { getRuntimeProviderProfile } from "@omniroute/open-sse/services/accountFallback.ts";
// Models that explicitly cannot run on the codex/ChatGPT-Pro OAuth pool — when
// a caller writes `codex/deepseek-v4-pro` we transparently reroute to the
// canonical provider whose API key is configured. Saves callers from having
// to know about the OAuth-vs-API-key split.
const NON_OAUTH_MODEL_PREFIX = /^(deepseek|qwen|kimi|glm|minimax|mimo)/i;
const PREFERRED_BY_FAMILY: Record<string, string> = {
deepseek: "deepseek",
qwen: "bailian",
kimi: "moonshot",
glm: "zhipu",
minimax: "minimax",
mimo: "moonshot",
};
export async function resolveModelOrError(modelStr: string, body: any, endpointPath: string = "") {
const modelInfo = await getModelInfo(modelStr);
// Forced-rewrite: codex provider doesn't serve DeepSeek/Qwen/Kimi/etc. Reroute
// these to their canonical native provider so the request lands on the right
// upstream API key instead of failing with a 400 on the OAuth account.
// Ambiguous candidates (e.g. deepseek-v4-pro lives on both ds + opencode-go)
// resolve to the model-family's native provider via NON_OAUTH_PROVIDER_BY_FAMILY.
if (
modelInfo.provider === "codex" &&
typeof modelInfo.model === "string" &&
NON_OAUTH_MODEL_PREFIX.test(modelInfo.model)
) {
log.info(
"ROUTING",
`codex/${modelInfo.model} → re-resolving via native provider (codex OAuth does not serve this model)`
);
const rerouted = await getModelInfo(modelInfo.model);
if (rerouted.provider && rerouted.provider !== "codex") {
log.info("ROUTING", `codex/${modelInfo.model} → ${rerouted.provider}/${rerouted.model}`);
Object.assign(modelInfo, rerouted);
} else if ((rerouted as any).errorType === "ambiguous_model") {
const candidates: string[] = (rerouted as any).candidateProviders || [];
const family = modelInfo.model.match(NON_OAUTH_MODEL_PREFIX)?.[1]?.toLowerCase();
const pick = family && PREFERRED_BY_FAMILY[family];
if (pick && candidates.includes(pick)) {
log.info(
"ROUTING",
`codex/${modelInfo.model} → ${pick}/${modelInfo.model} (ambiguity resolved by family)`
);
modelInfo.provider = pick;
modelInfo.model = (rerouted as any).model;
}
}
}
if (!modelInfo.provider) {
if ((modelInfo as any).errorType === "ambiguous_model") {
const message =
(modelInfo as any).errorMessage ||
`Ambiguous model '${modelStr}'. Use provider/model prefix (ex: gh/${modelStr} or cc/${modelStr}).`;
log.warn("CHAT", message, {
model: modelStr,
candidates:
(modelInfo as any).candidateAliases || (modelInfo as any).candidateProviders || [],
});
return { error: errorResponse(HTTP_STATUS.BAD_REQUEST, message) };
// Family disambiguation: if the model name begins with a known
// non-OAuth family prefix, auto-pick the family-native provider
// from the candidate set instead of returning a 400. Saves callers
// (codex CLI, hermes, etc.) from having to guess the right alias.
const candidates: string[] = (modelInfo as any).candidateProviders || [];
const modelLower = (modelInfo.model || modelStr).toLowerCase();
const family = modelLower.match(NON_OAUTH_MODEL_PREFIX)?.[1];
const pick = family && PREFERRED_BY_FAMILY[family];
if (pick && candidates.includes(pick)) {
log.info(
"ROUTING",
`${modelStr} → ${pick}/${modelInfo.model} (ambiguity auto-resolved by family)`
);
modelInfo.provider = pick;
} else {
const message =
(modelInfo as any).errorMessage ||
`Ambiguous model '${modelStr}'. Use provider/model prefix (ex: gh/${modelStr} or cc/${modelStr}).`;
log.warn("CHAT", message, {
model: modelStr,
candidates:
(modelInfo as any).candidateAliases || (modelInfo as any).candidateProviders || [],
});
return { error: errorResponse(HTTP_STATUS.BAD_REQUEST, message) };
}
} else {
log.warn("CHAT", "Invalid model format", { model: modelStr });
return { error: errorResponse(HTTP_STATUS.BAD_REQUEST, "Invalid model format") };
}
log.warn("CHAT", "Invalid model format", { model: modelStr });
return { error: errorResponse(HTTP_STATUS.BAD_REQUEST, "Invalid model format") };
}
const { provider, model, extendedContext } = modelInfo;