pi-freeflow 1.0.2 → 1.0.3

package/extensions/index.ts

@@ -844,6 +844,54 @@ function sanitizeHeaders(
 	return sanitized;
 }
 
+function normalizeRequestBody(
+	body: Record<string, unknown>,
+	isRelay = false,
+): Record<string, unknown> {
+	// 1. Tool choice normalization (OpenCode Zen only supports "auto" or undefined)
+	if (body.tool_choice === "none") {
+		delete body.tool_choice;
+		delete body.tools;
+	} else if (body.tool_choice && body.tool_choice !== "auto") {
+		body.tool_choice = "auto";
+	}
+
+	// 2. Reasoning normalization
+	if (typeof body.reasoning_effort === "string") {
+		const re = body.reasoning_effort.toLowerCase();
+		if (re === "xhigh" || re === "max") {
+			body.reasoning_effort = "max";
+		} else if (re === "high" || re === "medium") {
+			body.reasoning_effort = "high";
+		} else if (re === "none" || re === "off") {
+			delete body.reasoning_effort;
+		} else {
+			body.reasoning_effort = "low";
+		}
+	}
+	if (body.reasoning && typeof body.reasoning === "object") {
+		const r = body.reasoning as Record<string, unknown>;
+		if (r.effort === "none" || r.effort === "off") {
+			delete r.effort;
+		} else if (typeof r.effort === "string") {
+			const re = r.effort.toLowerCase();
+			if (re === "xhigh" || re === "max") r.effort = "max";
+			else if (re === "high" || re === "medium") r.effort = "high";
+			else r.effort = "low";
+		}
+	}
+
+	// 3. Max tokens clamp for Vercel relay
+	if (isRelay) {
+		const mt = body.max_tokens ?? body.maxTokens;
+		if (typeof mt === "number" && mt > RELAY_MAX_TOKENS) {
+			body.max_tokens = RELAY_MAX_TOKENS;
+		}
+	}
+
+	return body;
+}
+
 // ponytail: shared stream pipe — upstream abort/timeout must end response,
 // not become an uncaught exception that crashes pi.
 function pipeUpstreamStream(
@@ -1017,27 +1065,11 @@ function startProxy(
 							activeHost,
 						);
 						try {
-							// ponytail: clamp max_tokens for Vercel relay — large values
-							// cause 400 "Upstream request failed" (response size / duration
-							// limits). Direct mode stays unconstrained.
 							let relayBody = bodyChunks.length
 								? Buffer.concat(bodyChunks)
 								: undefined;
 							if (relayBody && parsedBody) {
-								const mt = parsedBody.max_tokens ?? parsedBody.maxTokens;
-								if (typeof mt === "number" && mt > RELAY_MAX_TOKENS) {
-									parsedBody.max_tokens = RELAY_MAX_TOKENS;
-								}
-								if (typeof parsedBody.reasoning_effort === "string") {
-									const re = parsedBody.reasoning_effort.toLowerCase();
-									if (re === "xhigh" || re === "max") {
-										parsedBody.reasoning_effort = "max";
-									} else if (re === "high" || re === "medium") {
-										parsedBody.reasoning_effort = "high";
-									} else {
-										parsedBody.reasoning_effort = "low";
-									}
-								}
+								normalizeRequestBody(parsedBody, true);
 								relayBody = Buffer.from(JSON.stringify(parsedBody));
 							}
 							const response = await relayFetch(fullUrl, {
@@ -1081,15 +1113,8 @@ function startProxy(
 					}
 					// direct path (existing, untouched)
 					let directBody = Buffer.concat(bodyChunks);
-					if (parsedBody && typeof parsedBody.reasoning_effort === "string") {
-						const re = parsedBody.reasoning_effort.toLowerCase();
-						if (re === "xhigh" || re === "max") {
-							parsedBody.reasoning_effort = "max";
-						} else if (re === "high" || re === "medium") {
-							parsedBody.reasoning_effort = "high";
-						} else {
-							parsedBody.reasoning_effort = "low";
-						}
+					if (parsedBody) {
+						normalizeRequestBody(parsedBody, false);
 						directBody = Buffer.from(JSON.stringify(parsedBody));
 					}
 					const fwd = sanitizeHeaders(req.headers, target.hostname);

package/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "pi-freeflow",
-	"version": "1.0.2",
+	"version": "1.0.3",
 	"description": "Personal multi-cloud rolling fallback relay for OpenCode Zen and KiloCode models in OMP/Pi",
 	"keywords": [
 		"pi-package",