feat: configurable tool name prefix (#199) and custom rpm/tpm rate limits (#198)

- Issue #199: proxy_ prefix on tool names is now automatically disabled
  when routing to non-Claude backends (Gemini, Compatible, etc.).
  Prevents tool name mismatches in OpenCode, Cursor, and other clients.

- Issue #198: Added customRpm/customTpm support per provider connection.
  Users can configure custom rate limits via connection settings,
  overriding the default auto-learned limits from response headers.

open-sse/handlers/chatCore.ts

@@ -151,11 +151,18 @@ export async function handleChatCore({
   // Translate request (pass reqLogger for intermediate logging)
   let translatedBody = body;
   try {
+    // Issue #199: Disable tool name prefix when routing Claude-format requests
+    // to non-Claude backends (prefix causes tool name mismatches)
+    const claudeProviders = ["claude", "anthropic"];
+    if (targetFormat === FORMATS.CLAUDE && !claudeProviders.includes(provider?.toLowerCase?.())) {
+      translatedBody = { ...translatedBody, _disableToolPrefix: true };
+    }
+
     translatedBody = translateRequest(
       sourceFormat,
       targetFormat,
       model,
-      body,
+      translatedBody,
       stream,
       credentials,
       provider,
@@ -202,6 +209,7 @@ export async function handleChatCore({
   // Extract toolNameMap for response translation (Claude OAuth)
   const toolNameMap = translatedBody._toolNameMap;
   delete translatedBody._toolNameMap;
+  delete translatedBody._disableToolPrefix;
 
   // Update model in body
   translatedBody.model = model;

open-sse/services/rateLimitManager.ts

@@ -81,6 +81,7 @@ export async function initializeRateLimits() {
     const connections = await getProviderConnections();
     let explicitCount = 0;
     let autoCount = 0;
+    let customCount = 0;
 
     for (const connRaw of connections as unknown[]) {
       const conn = toRecord(connRaw);
@@ -88,9 +89,33 @@ export async function initializeRateLimits() {
       const provider = typeof conn.provider === "string" ? conn.provider : "";
       const isActive = conn.isActive === true;
       const rateLimitProtection = conn.rateLimitProtection === true;
+      const customRpm = toNumber(conn.customRpm, 0);
+      const customTpm = toNumber(conn.customTpm, 0);
       if (!connectionId || !provider) continue;
 
-      if (rateLimitProtection) {
+      // Custom rpm/tpm configured — enable rate limiting with user-defined values (#198)
+      if (customRpm > 0 || customTpm > 0) {
+        enabledConnections.add(connectionId);
+        customCount++;
+
+        const key = `${provider}:${connectionId}`;
+        const rpm = customRpm > 0 ? customRpm : DEFAULT_API_LIMITS.requestsPerMinute;
+        const minTime = Math.max(0, Math.floor(60000 / rpm) - 10);
+
+        if (!limiters.has(key)) {
+          limiters.set(
+            key,
+            new Bottleneck({
+              maxConcurrent: DEFAULT_API_LIMITS.concurrentRequests,
+              minTime,
+              reservoir: rpm,
+              reservoirRefreshAmount: rpm,
+              reservoirRefreshInterval: 60 * 1000,
+              id: key,
+            })
+          );
+        }
+      } else if (rateLimitProtection) {
         // Explicitly enabled by user
         enabledConnections.add(connectionId);
         explicitCount++;
@@ -117,9 +142,9 @@ export async function initializeRateLimits() {
       }
     }
 
-    if (explicitCount > 0 || autoCount > 0) {
+    if (explicitCount > 0 || autoCount > 0 || customCount > 0) {
       console.log(
-        `🛡️ [RATE-LIMIT] Loaded ${explicitCount} explicit + ${autoCount} auto-enabled (API key) protection(s)`
+        `🛡️ [RATE-LIMIT] Loaded ${explicitCount} explicit + ${autoCount} auto-enabled + ${customCount} custom rpm/tpm protection(s)`
       );
     }
 

open-sse/translator/request/openai-to-claude.ts

@@ -5,7 +5,8 @@ import { adjustMaxTokens } from "../helpers/maxTokensHelper.ts";
 import { DEFAULT_THINKING_CLAUDE_SIGNATURE } from "../../config/defaultThinkingSignature.ts";
 
 // Prefix for Claude OAuth tool names to avoid conflicts
-const CLAUDE_OAUTH_TOOL_PREFIX = "proxy_";
+// Can be disabled per-request via body._disableToolPrefix = true
+export const CLAUDE_OAUTH_TOOL_PREFIX = "proxy_";
 const CLAUDE_TOOL_CHOICE_REQUIRED = "an" + "y";
 
 type ClaudeContentBlock = Record<string, unknown>;
@@ -27,6 +28,9 @@ type ClaudeTool = {
 
 // Convert OpenAI request to Claude format
 export function openaiToClaudeRequest(model, body, stream) {
+  // Check if tool prefix should be disabled (configured per-provider or global)
+  const disableToolPrefix = body?._disableToolPrefix === true;
+
   // Tool name mapping for Claude OAuth (capitalizedName → originalName)
   const toolNameMap = new Map();
   const result: {
@@ -82,7 +86,7 @@ export function openaiToClaudeRequest(model, body, stream) {
 
     for (const msg of nonSystemMessages) {
       const newRole = msg.role === "user" || msg.role === "tool" ? "user" : "assistant";
-      const blocks = getContentBlocksFromMessage(msg, toolNameMap);
+      const blocks = getContentBlocksFromMessage(msg, toolNameMap, disableToolPrefix);
       const hasToolUse = blocks.some((b) => b.type === "tool_use");
       const hasToolResult = blocks.some((b) => b.type === "tool_result");
 
@@ -155,10 +159,13 @@ export function openaiToClaudeRequest(model, body, stream) {
       const originalName = toolData.name;
 
       // Claude OAuth requires prefixed tool names to avoid conflicts
-      const toolName = CLAUDE_OAUTH_TOOL_PREFIX + originalName;
+      // When prefix is disabled (non-Claude backends), use original name
+      const toolName = disableToolPrefix ? originalName : CLAUDE_OAUTH_TOOL_PREFIX + originalName;
 
       // Store mapping for response translation (prefixed → original)
-      toolNameMap.set(toolName, originalName);
+      if (!disableToolPrefix) {
+        toolNameMap.set(toolName, originalName);
+      }
 
       return {
         name: toolName,
@@ -196,7 +203,7 @@ export function openaiToClaudeRequest(model, body, stream) {
 }
 
 // Get content blocks from single message
-function getContentBlocksFromMessage(msg, toolNameMap = new Map()) {
+function getContentBlocksFromMessage(msg, toolNameMap = new Map(), disableToolPrefix = false) {
   const blocks = [];
 
   if (msg.role === "tool") {
@@ -277,8 +284,8 @@ function getContentBlocksFromMessage(msg, toolNameMap = new Map()) {
           const fnName = tc.function?.name;
           if (!fnName || !fnName.trim()) continue;
 
-          // Apply prefix to tool name
-          const toolName = CLAUDE_OAUTH_TOOL_PREFIX + fnName;
+          // Apply prefix to tool name (skip if disabled)
+          const toolName = disableToolPrefix ? fnName : CLAUDE_OAUTH_TOOL_PREFIX + fnName;
           blocks.push({
             type: "tool_use",
             id: tc.id,

open-sse/translator/response/openai-to-claude.ts

@@ -1,8 +1,6 @@
 import { register } from "../registry.ts";
 import { FORMATS } from "../formats.ts";
-
-// Prefix for Claude OAuth tool names (must match request translator)
-const CLAUDE_OAUTH_TOOL_PREFIX = "proxy_";
+import { CLAUDE_OAUTH_TOOL_PREFIX } from "../request/openai-to-claude.ts";
 
 // Helper: stop thinking block if started
 function stopThinkingBlock(state, results) {