jasper-context-compactor 0.3.6 → 0.3.8

package/README.md

@@ -33,12 +33,13 @@ npx jasper-context-compactor setup
 ### Supported Local Providers
 
 The setup automatically detects these providers (primary or fallback):
-- **Ollama** — `/ollama` endpoint
+- **Ollama** — Any provider with `ollama` in name or `:11434` in baseUrl
 - **llama.cpp** — llamacpp provider
 - **MLX** — mlx provider  
 - **LM Studio** — lmstudio provider
 - **friend-gpu** — Custom GPU servers
 - **OpenRouter** — When routing to local models
+- **Local network** — Any provider with localhost, 127.0.0.1, or Tailscale IP in baseUrl
 
 Then restart OpenClaw:
 ```bash
@@ -81,7 +82,8 @@ The setup configures these values in `~/.openclaw/openclaw.json`:
           "maxTokens": 8000,
           "keepRecentTokens": 2000,
           "summaryMaxTokens": 1000,
-          "charsPerToken": 4
+          "charsPerToken": 4,
+          "modelFilter": ["ollama", "lmstudio"]
         }
       }
     }
@@ -95,6 +97,7 @@ The setup configures these values in `~/.openclaw/openclaw.json`:
 | `keepRecentTokens` | Recent context to preserve (default: 25% of max) |
 | `summaryMaxTokens` | Max tokens for the summary (default: 12.5% of max) |
 | `charsPerToken` | Token estimation ratio (4 works for English) |
+| `modelFilter` | (Optional) Only compact for these providers. If not set, compacts all sessions.
 
 ## Restoring Your Config
 

package/cli.js

@@ -48,12 +48,42 @@ function backupConfig() {
 }
 
 // Local model providers that benefit from context compaction
-const LOCAL_PROVIDERS = ['ollama', 'lmstudio', 'llamacpp', 'mlx', 'friend-gpu', 'openrouter'];
+const LOCAL_PROVIDER_NAMES = ['ollama', 'lmstudio', 'llamacpp', 'mlx', 'friend-gpu', 'openrouter'];
 
-function isLocalProvider(providerId) {
+// URLs that indicate local/Ollama endpoints
+const LOCAL_URL_PATTERNS = [
+  ':11434',           // Ollama default port
+  'localhost',
+  '127.0.0.1',
+  '0.0.0.0',
+  /100\.\d+\.\d+\.\d+/,  // Tailscale
+  /192\.168\.\d+\.\d+/,  // Local network
+  /10\.\d+\.\d+\.\d+/,   // Private network
+];
+
+function isLocalProvider(providerId, providers = {}) {
   if (!providerId) return false;
   const lower = providerId.toLowerCase();
-  return LOCAL_PROVIDERS.some(p => lower.includes(p));
+  
+  // Check provider name
+  if (LOCAL_PROVIDER_NAMES.some(p => lower.includes(p))) {
+    return true;
+  }
+  
+  // Check provider's baseUrl for local patterns
+  const provider = providers[providerId];
+  if (provider?.baseUrl) {
+    const url = provider.baseUrl.toLowerCase();
+    for (const pattern of LOCAL_URL_PATTERNS) {
+      if (pattern instanceof RegExp) {
+        if (pattern.test(url)) return true;
+      } else {
+        if (url.includes(pattern)) return true;
+      }
+    }
+  }
+  
+  return false;
 }
 
 async function detectModelContextWindow(config) {
@@ -78,8 +108,8 @@ async function detectModelContextWindow(config) {
     const [providerName, ...modelParts] = modelId.split('/');
     const modelName = modelParts.join('/'); // e.g., "qwen2.5"
     
-    // Check if this is a local provider
-    if (isLocalProvider(providerName)) {
+    // Check if this is a local provider (by name or baseUrl)
+    if (isLocalProvider(providerName, providers)) {
       hasLocalModel = true;
       
       const provider = providers[providerName];
@@ -251,15 +281,6 @@ async function setup() {
       console.log('  Local models (Ollama, llama.cpp, MLX, LM Studio) don\'t report');
       console.log('  context overflow errors — they silently truncate or produce garbage.');
       console.log('  This plugin is HIGHLY recommended for your setup.');
-    } else if (detectedInfo?.model) {
-      // Cloud-only config
-      const providerName = detectedInfo.model.split('/')[0] || '';
-      if (['anthropic', 'openai', 'google'].includes(providerName.toLowerCase())) {
-        console.log('');
-        console.log('  ℹ️  Cloud-only config detected (no local models).');
-        console.log('     Cloud APIs report context limits properly, so this plugin');
-        console.log('     is less critical — but can still help with token costs.');
-      }
     }
     
     if (detectedInfo && detectedInfo.tokens) {

package/index.ts

@@ -24,6 +24,12 @@ interface PluginConfig {
   charsPerToken?: number;
   summaryModel?: string;
   logLevel?: 'debug' | 'info' | 'warn' | 'error';
+  /**
+   * Optional: Only run compaction when session model matches these providers.
+   * Example: ['ollama', 'lmstudio', 'friend-gpu']
+   * If not set, compaction runs for all sessions when enabled.
+   */
+  modelFilter?: string[];
 }
 
 interface Message {
@@ -162,8 +168,9 @@ export default function register(api: PluginApi) {
   const summaryMaxTokens = cfg.summaryMaxTokens ?? 1000;
   const charsPerToken = cfg.charsPerToken ?? 4;
   const summaryModel = cfg.summaryModel;
+  const modelFilter = cfg.modelFilter; // Optional: ['ollama', 'lmstudio', etc.]
 
-  api.logger.info(`[context-compactor] Initialized (maxTokens=${maxTokens}, keepRecent=${keepRecentTokens})`);
+  api.logger.info(`[context-compactor] Initialized (maxTokens=${maxTokens}, keepRecent=${keepRecentTokens}${modelFilter ? `, filter=${modelFilter.join(',')}` : ''})`);
 
   // ============================================================================
   // Core: before_agent_start hook
@@ -173,12 +180,23 @@ export default function register(api: PluginApi) {
     prompt?: string;
     sessionKey?: string;
     sessionId?: string;
+    model?: string;
     context?: {
       sessionFile?: string;
       messages?: Message[];
     };
   }) => {
     try {
+      // If modelFilter is set, only run for matching providers
+      if (modelFilter && modelFilter.length > 0 && event.model) {
+        const modelLower = event.model.toLowerCase();
+        const matches = modelFilter.some(filter => modelLower.includes(filter.toLowerCase()));
+        if (!matches) {
+          api.logger.debug?.(`[context-compactor] Skipping - model ${event.model} not in filter`);
+          return;
+        }
+      }
+
       // Get current messages from context or session file
       let messages: Message[] = event.context?.messages ?? [];
       

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "jasper-context-compactor",
-  "version": "0.3.6",
+  "version": "0.3.8",
   "description": "Context compaction plugin for OpenClaw - works with local models (MLX, llama.cpp) that don't report token limits",
   "main": "index.ts",
   "bin": {