Never stop coding. Free MIT AI gateway: one endpoint, 375 providers (150+ free), 1,400+ models Kimi, Claude, GPT, Gemini, GLM, DeepSeek, MiniMax. Works with Claude Code, Codex, Cursor, OpenCode, Cline & Copilot. Quota-aware auto-fallback, RTK+Caveman compression saves 15-95% tokens, MCP/A2A, Desktop/PWA. Built by 900+ contributors
TypeScript a2aai-agentsai-gatewayanthropicclaudeclaude-codeclinecodexcopilotcursordeepseekfree-aigeminikimillm-gatewaymcpopenaiopenai-proxyqwentoken-saver
import type { LanguageModelV3 } from "@ai-sdk/provider";
import { type Logger, isGeminiModelId, sanitizeToolInputSchemas } from "./shared/index.js";
type CallOptions = Parameters<LanguageModelV3["doGenerate"]>[0];
/**
* Gemini answers `400 INVALID_ARGUMENT` — for the entire request, not just the
* offending tool — when a tool declaration carries `$schema` or
* `additionalProperties`. Anything upstream that emits standard JSON Schema
* therefore breaks tool calling as soon as the chain routes to Gemini. A
* `$ref` is forwarded untouched instead: stripping it would widen the schema
* to "accept anything", which is worse than letting the gateway answer. The
* v1 plugin dealt with this by wrapping `fetch` and rewriting the JSON body; the
* v2 home for it is the language model, where the tools are still structured
* data and no re-parsing is needed.
*
* Returns the model untouched when it is not bound for Gemini, so the wrapper
* costs nothing on every other chain.
*/
export function sanitizeToolSchemasFor<T extends LanguageModelV3 | undefined>(
language: T,
modelId: string,
log: Logger
): T {
if (language === undefined) return language;
if (!isGeminiModelId(modelId)) return language;
const clean = (options: CallOptions): CallOptions => {
const tools = sanitizeToolInputSchemas(options.tools);
if (tools === undefined) return options;
log.debug(
`[omniroute-v2] stripped Gemini-incompatible schema keywords from ${tools.length} tool declaration(s) for ${modelId}`
);
return { ...options, tools } as CallOptions;
};
// Prototype-linked so every other member of the model — including accessors
// and anything a future SDK version adds — keeps working untouched.
const wrapped: LanguageModelV3 = Object.create(language as object) as LanguageModelV3;
wrapped.doGenerate = (options) => language.doGenerate(clean(options));
wrapped.doStream = (options) => language.doStream(clean(options));
return wrapped as T;
}