@quantish/agent 0.1.22 → 0.1.24

package/README.md

@@ -1,116 +1,105 @@
-# @quantish/cli
+# @quantish/agent
 
-AI-powered CLI agent for building trading bots on Polymarket.
+AI-powered coding & trading agent for Polymarket. Build trading bots, analyze markets, and execute trades using natural language.
 
-Combines **coding tools** (file system, shell, git) with **trading tools** (Polymarket orders, positions, wallet) powered by Claude AI.
+## ✨ Features
 
-## How It Works
-
-Quantish CLI connects to the **Quantish Signing Server** to execute trades on Polymarket. Here's why:
-
-### Why We Use a Signing Server
-
-Polymarket uses a **gasless relayer system** - this means:
-- ✅ **Free wallet creation** - No MATIC needed to set up
-- ✅ **Free trading** - Polymarket covers gas fees on all transactions
-- ✅ **Simplified signing** - Our server handles the complex signature formats
-
-To make this work reliably, the Quantish Signing Server:
-1. **Handles wallet creation** - Your wallet is created and managed through our server
-2. **Signs transactions** - Orders are signed using Polymarket's required format
-3. **Relays to Polymarket** - Transactions go through Polymarket's official relayer
-4. **Bypasses geo-restrictions** - Our server is hosted in a compatible region
-
-### What This Means for You
-
-- **Your funds are secure** - Only you can authorize transactions via your API key
-- **Wallets are non-custodial** - You can export your private key anytime with `export_private_key`
-- **Trading is free** - No gas fees, ever
-- **It just works** - No VPN or complex setup needed
-
-> 🔒 **Security Note**: Your private keys are stored encrypted. You can export them and migrate to a self-hosted solution in the future if needed.
+- **🤖 Multi-Provider AI** - Use Anthropic Claude or 100+ OpenRouter models (GLM-4.7, MiniMax, DeepSeek, etc.)
+- **💹 Live Trading** - Place orders, manage positions, check balances on Polymarket
+- **🔧 Full Coding Tools** - Read/write files, run commands, git operations
+- **🌐 Web Search** - Search the web with Exa AI or DuckDuckGo fallback
+- **💾 Session Persistence** - Save and resume conversations across sessions
+- **⚡ Queued Input** - Type while the agent is working, queue messages
+- **📊 Cost Tracking** - Real-time token usage and cost display
 
 ## Installation
 
 ```bash
-npm install -g @quantish/cli
-```
-
-Or run directly with npx:
-
-```bash
-npx @quantish/cli
+npm install -g @quantish/agent
 ```
 
 ## Quick Start
 
-### 1. Initialize
-
-Set up your API keys:
-
 ```bash
+# First-time setup
 quantish init
-```
-
-You'll need:
-- **Anthropic API Key** - Get one at https://console.anthropic.com/
-- **Quantish API Key** - Created automatically during setup
-
-### 2. Start Building
 
-**Interactive mode:**
-
-```bash
+# Start interactive chat
 quantish
 ```
 
-Example conversations:
+## How It Works
 
-```
-You: Create a trading bot that monitors my positions and sells when profit > 20%
-Assistant: I'll create that for you. Let me first check your current positions...
-[Calling get_positions...]
-[Writing bot.ts...]
+The agent connects to two MCP (Model Context Protocol) servers:
 
-You: What's my current balance?
-Assistant: Your Safe wallet has 0.68 USDC available for trading.
+1. **Discovery MCP** (Public) - Market search, trending markets, market details
+2. **Trading MCP** (Your API Key) - Wallet, orders, positions, trades
 
-You: Place a $5 YES order on Trump winning at 55 cents
-Assistant: Order placed! Order ID: abc123...
-```
+Your wallet is created and managed through our signing server, which:
+- ✅ Handles gasless transactions (Polymarket covers fees)
+- ✅ Signs orders using Polymarket's required format  
+- ✅ Works globally (no geo-restrictions)
+- 🔒 Non-custodial - export your private key anytime
 
-**One-shot mode:**
+## Interactive Commands
 
-```bash
-quantish -p "check my open orders"
-```
+### Chat Commands
+
+| Command | Description |
+|---------|-------------|
+| `/help` | Show all commands |
+| `/clear` | Clear conversation history |
+| `/compact` | Summarize conversation to save tokens |
+| `/model <name>` | Switch model (opus, sonnet, haiku, glm, minimax, etc.) |
+| `/provider <name>` | Switch LLM provider (anthropic, openrouter) |
+| `/cost` | Show session cost breakdown |
+| `/tools` | List available tools |
+| `/config` | Show configuration info |
 
-**Piped input:**
+### Session Commands
 
-```bash
-echo "show my positions" | quantish
-```
+| Command | Description |
+|---------|-------------|
+| `/save [name]` | Save current session |
+| `/resume` | Resume last session |
+| `/sessions` | List all saved sessions |
+| `/load <id>` | Load a session by ID |
+| `/forget` | Delete all saved sessions |
 
-## Commands
+### Process Commands
 
 | Command | Description |
 |---------|-------------|
-| `quantish` | Start interactive chat |
-| `quantish init` | Configure API keys |
-| `quantish config` | View configuration |
-| `quantish tools` | List available tools |
-| `quantish -p "..."` | Run one-shot prompt |
+| `/processes` | List running background processes |
+| `/stop <id>` | Stop a background process |
+| `/stopall` | Stop all background processes |
+
+### Keyboard Shortcuts
+
+| Key | Action |
+|-----|--------|
+| `Enter` | Send message (or queue if agent is working) |
+| `Esc` | Interrupt current generation |
+| `Ctrl+C` | Exit CLI |
 
-## Options
+## CLI Options
+
+```bash
+quantish                    # Interactive mode
+quantish init               # First-time setup wizard
+quantish config             # View configuration
+quantish config --export    # Export as .env format
+quantish tools              # List all available tools
+quantish -p "message"       # One-shot mode
+quantish --version          # Show version
+```
 
 | Option | Description |
 |--------|-------------|
 | `-p, --prompt <message>` | Run a single prompt |
-| `-v, --verbose` | Show tool calls |
+| `-v, --verbose` | Show detailed tool calls |
 | `--no-mcp` | Disable trading tools |
 | `--no-local` | Disable coding tools |
-| `--version` | Show version |
-| `--help` | Show help |
 
 ## Available Tools
 
@@ -118,118 +107,166 @@ echo "show my positions" | quantish
 
 | Tool | Description |
 |------|-------------|
-| `read_file` | Read file contents |
-| `write_file` | Write/create files |
+| `read_file` | Read file contents with line numbers |
+| `write_file` | Create or overwrite files |
+| `edit_file` | Search and replace in files |
+| `edit_lines` | Edit specific line ranges (efficient) |
 | `list_dir` | List directory contents |
 | `delete_file` | Delete files |
 | `file_exists` | Check if file exists |
-| `run_command` | Execute shell commands |
+| `run_command` | Execute shell commands (blocking) |
+| `start_background_process` | Run long-running processes |
+| `get_process_output` | Get output from background process |
+| `stop_process` | Stop a background process |
 | `grep` | Search file contents |
 | `find_files` | Find files by pattern |
-| `git_status` | Get git status |
-| `git_diff` | Show git diff |
+| `setup_env` | Create/update .env files |
+
+### Git Tools
+
+| Tool | Description |
+|------|-------------|
+| `git_status` | Get repository status |
+| `git_diff` | Show changes |
 | `git_add` | Stage files |
 | `git_commit` | Create commits |
 | `git_log` | Show commit history |
 | `git_checkout` | Switch branches |
-| `web_search` | Search the web (Exa/DuckDuckGo) |
-| `web_answer` | AI-powered Q&A (Exa) |
-| `fetch_url` | Fetch URL content |
 
-### MCP Tools (Trading)
+### Web Tools
 
 | Tool | Description |
 |------|-------------|
-| `get_balances` | Check wallet balances |
-| `get_positions` | View current positions |
-| `place_order` | Place buy/sell orders |
-| `cancel_order` | Cancel open orders |
-| `get_orders` | List orders |
-| `get_orderbook` | Get market orderbook |
-| `get_price` | Get current price |
-| `transfer_usdc` | Transfer USDC |
-| `swap_tokens` | Swap tokens |
-| `claim_winnings` | Claim from resolved markets |
-
-## Configuration
+| `web_search` | Search the web (Exa/DuckDuckGo) |
+| `web_answer` | AI-powered Q&A (requires Exa API key) |
+| `fetch_url` | Fetch URL content |
 
-Configuration is stored in `~/.quantish/config.json`:
+### MCP Tools (Trading)
 
-```json
-{
-  "anthropicApiKey": "sk-ant-...",
-  "quantishApiKey": "pk_live_...",
-  "mcpServerUrl": "https://quantish-sdk-production.up.railway.app/mcp"
-}
-```
+| Tool | Server | Description |
+|------|--------|-------------|
+| `search_markets` | Discovery | Search markets by query |
+| `get_trending_markets` | Discovery | Get trending/popular markets |
+| `get_market_details` | Discovery | Get market info and prices |
+| `get_balances` | Trading | Check wallet balances |
+| `get_positions` | Trading | View current positions |
+| `place_order` | Trading | Place buy/sell orders |
+| `cancel_order` | Trading | Cancel open orders |
+| `get_orders` | Trading | List orders |
+| `get_orderbook` | Trading | Get market orderbook |
+| `get_price` | Trading | Get current price |
+| `transfer_usdc` | Trading | Transfer USDC |
+| `claim_winnings` | Trading | Claim from resolved markets |
+| `export_private_key` | Trading | Export wallet private key |
 
-Environment variables take precedence:
-- `ANTHROPIC_API_KEY`
-- `QUANTISH_API_KEY`
+## LLM Providers
 
-## Examples
+### Anthropic (Default for new installs)
 
-### Build a Trading Bot
+Uses Claude models directly via Anthropic API.
 
 ```bash
-quantish
-> Create a Python script that monitors the Trump market and alerts me when price drops below 40 cents
+/model opus    # Claude Opus 4.5 - Most capable
+/model sonnet  # Claude Sonnet 4.5 - Balanced (default)
+/model haiku   # Claude Haiku 4.5 - Fastest/cheapest
 ```
 
-### Manage Positions
-
-```bash
-quantish -p "show me my positions with unrealized P&L"
-```
+### OpenRouter
 
-### Market Making
+Access 100+ models from various providers.
 
 ```bash
-quantish
-> Help me set up a basic market making strategy. I want to place both bid and ask orders around the current mid price.
-```
+/provider openrouter  # Switch to OpenRouter
 
-### Code Review
+/model glm      # GLM-4.7 (default for OpenRouter) - Best for coding
+/model minimax  # MiniMax M2.1 - Fast and cheap
+/model deepseek # DeepSeek V3.2 - Great reasoning
+/model gemini   # Gemini 2.0 Flash - Google's latest
+/model grok     # Grok 3 Mini Beta - xAI
+```
 
+Or use any OpenRouter model ID:
 ```bash
-quantish
-> Read my trading bot code in bot.ts and suggest improvements
+/model anthropic/claude-3.5-sonnet
+/model meta-llama/llama-3.3-70b-instruct
 ```
 
-## Development
+## Configuration
 
-```bash
-# Clone the repo
-git clone https://github.com/quantish/cli
+Configuration is stored in `~/.quantish/config.json`.
 
-# Install dependencies
-cd packages/quantish-cli
-npm install
+### Environment Variables
 
-# Build
-npm run build
+| Variable | Description |
+|----------|-------------|
+| `ANTHROPIC_API_KEY` | Anthropic API key |
+| `OPENROUTER_API_KEY` | OpenRouter API key |
+| `QUANTISH_API_KEY` | Quantish trading API key |
+| `EXA_API_KEY` | Exa AI search key (optional) |
+| `MCP_SERVER_URL` | Custom Trading MCP server URL |
 
-# Run locally
-npm start
+### Export Configuration
 
-# Development mode (watch)
-npm run dev
+```bash
+quantish config --export > .env
 ```
 
-## Architecture
-
-```
-quantish (CLI)
-    │
-    ├── Local Tools (filesystem, shell, git)
-    │   └── Runs directly on your machine
-    │
-    └── MCP Tools (trading)
-        └── Calls Quantish MCP Server
-            └── Executes on Polymarket
+## Building Applications
+
+The agent can build standalone applications that use the Quantish MCP API. When building apps, ensure:
+
+1. **Use HTTP API** - Don't use MCP SDK directly
+2. **Environment Variables** - Store API keys in `.env`
+3. **Two Endpoints**:
+   - Discovery: `https://quantish.live/mcp/execute` (public)
+   - Trading: `https://quantish-sdk-production.up.railway.app/mcp/execute` (requires API key)
+
+Example API call:
+
+```javascript
+// Discovery MCP (simple format)
+const response = await fetch('https://quantish.live/mcp/execute', {
+  method: 'POST',
+  headers: {
+    'Content-Type': 'application/json',
+    'X-API-Key': 'qm_ueQeqrmvZyHtR1zuVbLYkhx0fKyVAuV8'
+  },
+  body: JSON.stringify({
+    name: 'search_markets',
+    arguments: { query: 'bitcoin', limit: 5 }
+  })
+});
+
+// Trading MCP (JSON-RPC format)
+const response = await fetch('https://quantish-sdk-production.up.railway.app/mcp/execute', {
+  method: 'POST',
+  headers: {
+    'Content-Type': 'application/json',
+    'X-API-Key': process.env.QUANTISH_API_KEY
+  },
+  body: JSON.stringify({
+    jsonrpc: '2.0',
+    id: 1,
+    method: 'tools/call',
+    params: {
+      name: 'get_positions',
+      arguments: {}
+    }
+  })
+});
 ```
 
-The agent uses Claude to understand your requests and decide which tools to use. It can combine coding and trading tools in a single conversation.
+## Self-Hosting
+
+You can self-host the Quantish MCP server:
+
+```bash
+# Set custom server URL
+quantish config --server https://your-server.com/mcp
+
+# Or use environment variable
+export MCP_SERVER_URL=https://your-server.com/mcp
+```
 
 ## Platform Support
 
@@ -239,24 +276,71 @@ The agent uses Claude to understand your requests and decide which tools to use.
 | Linux | ✅ Full support |
 | Windows | ⚠️ Requires WSL |
 
-**Windows users:** Install [WSL (Windows Subsystem for Linux)](https://learn.microsoft.com/en-us/windows/wsl/install) and run Quantish from within WSL. Native Windows (PowerShell/cmd.exe) is not supported.
+## Examples
+
+```bash
+# Search for markets
+quantish -p "find markets about bitcoin"
 
-## Environment Variables
+# Check positions
+quantish -p "show my positions with P&L"
 
-| Variable | Description |
-|----------|-------------|
-| `ANTHROPIC_API_KEY` | Your Anthropic API key (required) |
-| `QUANTISH_API_KEY` | Your Quantish trading API key |
-| `EXA_API_KEY` | Optional: Exa AI search key for powerful web search |
+# Build a trading bot
+quantish
+> Create a bot that monitors Trump markets and alerts me when prices change more than 5%
+
+# Start a dev server
+quantish
+> Start my React app on port 3001
+
+# Code review
+quantish
+> Review my trading bot code and suggest improvements
+```
+
+## Troubleshooting
+
+### Tool calls failing with malformed arguments
+
+Some OpenRouter models (like GLM-4.7) occasionally emit malformed tool calls. The CLI includes robust parsing to handle these, but if issues persist:
+
+```bash
+/model sonnet  # Switch to Claude Sonnet
+```
+
+### Session not resuming
 
-### Web Search
+Sessions are stored in `~/.quantish/sessions/`. To reset:
 
-Web search works without API keys (using DuckDuckGo fallback), but **Exa is strongly recommended** for AI-quality search results.
+```bash
+rm -rf ~/.quantish/sessions
+```
 
-Get your Exa API key at: https://dashboard.exa.ai
+### High token usage
 
-Exa is the same search engine used by Cursor, Notion, Vercel, and other leading AI products.
+```bash
+/compact       # Summarize conversation
+/model haiku   # Switch to cheaper model
+/clear         # Start fresh
+```
+
+## Development
+
+```bash
+git clone https://github.com/joinQuantish/quantish-agent
+cd quantish-agent
+npm install
+npm run build
+npm link  # Install locally
+```
 
 ## License
 
 MIT
+
+## Links
+
+- [GitHub](https://github.com/joinQuantish/quantish-agent)
+- [NPM](https://www.npmjs.com/package/@quantish/agent)
+- [Documentation](https://docs.quantish.live)
+- [Quantish Platform](https://quantish.live)

package/dist/index.js

@@ -14,6 +14,8 @@ var DEFAULT_TRADING_MCP_URL = "https://quantish-sdk-production.up.railway.app/mc
 var DISCOVERY_MCP_URL = "https://quantish.live/mcp";
 var DISCOVERY_MCP_PUBLIC_KEY = "qm_ueQeqrmvZyHtR1zuVbLYkhx0fKyVAuV8";
 var DEFAULT_MCP_URL = DEFAULT_TRADING_MCP_URL;
+var DEFAULT_ANTHROPIC_MODEL = "claude-sonnet-4-5-20250929";
+var DEFAULT_OPENROUTER_MODEL = "z-ai/glm-4.7";
 var schema = {
   anthropicApiKey: {
     type: "string"
@@ -132,10 +134,13 @@ var ConfigManager = class {
     this.conf.set("mcpServerUrl", url);
   }
   /**
-   * Get the model to use
+   * Get the model to use (returns default based on current provider)
    */
   getModel() {
-    return this.conf.get("model") ?? "claude-sonnet-4-5-20250929";
+    const model = this.conf.get("model");
+    if (model) return model;
+    const provider = this.getProvider();
+    return provider === "openrouter" ? DEFAULT_OPENROUTER_MODEL : DEFAULT_ANTHROPIC_MODEL;
   }
   /**
    * Set the model to use
@@ -2208,114 +2213,6 @@ async function executeLocalTool(name, args) {
   return { success: false, error: `Unknown local tool: ${name}` };
 }
 
-// src/agent/compaction.ts
-var COMPACTION_PROMPT = `Your context window is filling up. Please create a concise summary of our conversation so far that will allow you to continue working effectively.
-
-The summary should be wrapped in <summary></summary> tags and include:
-
-# Task Overview
-- The user's core request and goals
-- Success criteria and constraints
-- Any specific preferences mentioned
-
-# Current State
-- What has been completed so far
-- Files created or modified (with paths)
-- Artifacts or outputs produced
-- Current working directory if relevant
-
-# Important Discoveries
-- Technical constraints or requirements found
-- Key decisions made and why
-- Errors encountered and how they were resolved
-- Approaches that didn't work (to avoid repeating)
-
-# Next Steps
-- Specific actions still needed
-- Priority order if multiple steps remain
-- Any blockers or dependencies
-
-# Context to Preserve
-- User preferences or style requirements
-- Domain-specific details that matter
-- Any commitments or promises made
-
-Be thorough but concise. The goal is to capture everything needed to continue seamlessly, while reducing token usage significantly.`;
-function parseCompactedSummary(response) {
-  const match = response.match(/<summary>([\s\S]*?)<\/summary>/);
-  if (match && match[1]) {
-    return match[1].trim();
-  }
-  return response.trim() || null;
-}
-async function createCompactedSummary(anthropic, history, model = "claude-sonnet-4-5-20250929", customPrompt) {
-  const prompt2 = customPrompt || COMPACTION_PROMPT;
-  const compactionMessages = [
-    ...history,
-    {
-      role: "user",
-      content: prompt2
-    }
-  ];
-  const response = await anthropic.messages.create({
-    model,
-    max_tokens: 4096,
-    messages: compactionMessages
-  });
-  const textBlocks = response.content.filter((block) => block.type === "text");
-  const fullText = textBlocks.map((block) => block.text).join("\n");
-  const summary = parseCompactedSummary(fullText);
-  if (!summary) {
-    throw new Error("Failed to parse compacted summary from response");
-  }
-  return summary;
-}
-function historyFromSummary(summary) {
-  return [
-    {
-      role: "assistant",
-      content: summary
-    }
-  ];
-}
-async function compactConversation(anthropic, history, model, systemPrompt, tools) {
-  let originalTokens = 0;
-  try {
-    const countResult = await anthropic.messages.countTokens({
-      model,
-      system: systemPrompt,
-      tools,
-      messages: history
-    });
-    originalTokens = countResult.input_tokens;
-  } catch (e) {
-    const contentLength = JSON.stringify(history).length;
-    originalTokens = Math.ceil(contentLength / 4);
-  }
-  const summaryModel = "claude-sonnet-4-5-20250929";
-  const summary = await createCompactedSummary(anthropic, history, summaryModel);
-  const newHistory = historyFromSummary(summary);
-  let newTokens = 0;
-  try {
-    const countResult = await anthropic.messages.countTokens({
-      model,
-      system: systemPrompt,
-      tools,
-      messages: newHistory
-    });
-    newTokens = countResult.input_tokens;
-  } catch (e) {
-    const contentLength = JSON.stringify(newHistory).length;
-    newTokens = Math.ceil(contentLength / 4);
-  }
-  return {
-    newHistory,
-    summary,
-    originalTokens,
-    newTokens
-  };
-}
-
 // src/agent/pricing.ts
 var MODELS = {
   "claude-opus-4-5-20250929": {
@@ -2937,17 +2834,33 @@ var OpenRouterClient = class {
   }
 };
 function calculateOpenRouterCost(modelId, inputTokens, outputTokens, cacheReadTokens = 0, cacheWriteTokens = 0) {
-  const config = getOpenRouterModelConfig(modelId);
+  let config = getOpenRouterModelConfig(modelId);
+  if (!config) {
+    config = getOpenRouterModelConfig(modelId.toLowerCase());
+  }
+  if (!config) {
+    const lower = modelId.toLowerCase();
+    for (const [key, model] of Object.entries(OPENROUTER_MODELS2)) {
+      if (key.toLowerCase() === lower || model.name.toLowerCase() === lower) {
+        config = model;
+        break;
+      }
+    }
+    if (!config && OPENROUTER_ALIASES[lower]) {
+      config = OPENROUTER_MODELS2[OPENROUTER_ALIASES[lower]];
+    }
+  }
   const pricing = config?.pricing ?? {
-    inputPerMTok: 1,
-    outputPerMTok: 3,
-    cacheReadPerMTok: 0.1,
-    cacheWritePerMTok: 1.25
+    inputPerMTok: 0.4,
+    // GLM 4.7 pricing as fallback
+    outputPerMTok: 1.5,
+    cacheReadPerMTok: 0,
+    cacheWritePerMTok: 0
   };
   const inputCost = inputTokens / 1e6 * pricing.inputPerMTok;
   const outputCost = outputTokens / 1e6 * pricing.outputPerMTok;
-  const cacheReadCost = cacheReadTokens / 1e6 * (pricing.cacheReadPerMTok ?? pricing.inputPerMTok * 0.1);
-  const cacheWriteCost = cacheWriteTokens / 1e6 * (pricing.cacheWritePerMTok ?? pricing.inputPerMTok * 1.25);
+  const cacheReadCost = cacheReadTokens / 1e6 * (pricing.cacheReadPerMTok ?? 0);
+  const cacheWriteCost = cacheWriteTokens / 1e6 * (pricing.cacheWritePerMTok ?? 0);
   return {
     inputCost,
     outputCost,
@@ -3264,7 +3177,34 @@ var OpenRouterProvider = class {
         if (!tc || !tc.name) {
           continue;
         }
+        let toolName = tc.name;
+        if (toolName.includes("<")) {
+          toolName = toolName.split("<")[0];
+        }
+        if (toolName.includes("(")) {
+          toolName = toolName.split("(")[0];
+        }
+        toolName = toolName.trim();
         let args = tc.arguments?.trim() || "{}";
+        if (args.includes("<arg_key>") || args.includes("</arg_key>")) {
+          args = args.replace(/<\/?arg_key>/g, "");
+          if (!args.startsWith("{")) {
+            const keyValuePairs = [];
+            const kvMatches = args.matchAll(/(\w+):\s*(?:"([^"]+)"|(\d+)|(\w+))/g);
+            for (const match of kvMatches) {
+              const key = match[1];
+              const value = match[2] ?? match[3] ?? match[4];
+              if (match[3]) {
+                keyValuePairs.push(`"${key}": ${value}`);
+              } else {
+                keyValuePairs.push(`"${key}": "${value}"`);
+              }
+            }
+            if (keyValuePairs.length > 0) {
+              args = `{${keyValuePairs.join(", ")}}`;
+            }
+          }
+        }
         if (args && !args.endsWith("}") && !args.endsWith("]")) {
           const openBraces = (args.match(/{/g) || []).length;
           const closeBraces = (args.match(/}/g) || []).length;
@@ -3277,10 +3217,10 @@ var OpenRouterProvider = class {
         }
         const input = JSON.parse(args);
         const toolId = tc.id || `tool_${Date.now()}_${Math.random().toString(36).slice(2)}`;
-        toolCalls.push({ id: toolId, name: tc.name, input });
-        callbacks.onToolCall?.(toolId, tc.name, input);
+        toolCalls.push({ id: toolId, name: toolName, input });
+        callbacks.onToolCall?.(toolId, toolName, input);
       } catch (e) {
-        const toolName = tc?.name || "unknown_tool";
+        const cleanToolName = tc?.name?.split("<")[0]?.split("(")[0]?.trim() || "unknown_tool";
         let parsedInput = {};
         try {
           const argsStr = tc?.arguments || "";
@@ -3291,8 +3231,8 @@ var OpenRouterProvider = class {
         } catch {
         }
         const toolId = tc?.id || `tool_${Date.now()}_${Math.random().toString(36).slice(2)}`;
-        toolCalls.push({ id: toolId, name: toolName, input: parsedInput });
-        callbacks.onToolCall?.(toolId, toolName, parsedInput);
+        toolCalls.push({ id: toolId, name: cleanToolName, input: parsedInput });
+        callbacks.onToolCall?.(toolId, cleanToolName, parsedInput);
       }
     }
     const cost = calculateOpenRouterCost(
@@ -3689,7 +3629,7 @@ ${userMessage}`;
         output_tokens: response.usage.outputTokens,
         cache_creation_input_tokens: response.usage.cacheCreationTokens,
         cache_read_input_tokens: response.usage.cacheReadTokens
-      });
+      }, response.cost);
       const responseContent = [];
       if (response.text) {
         responseContent.push({ type: "text", text: response.text });
@@ -4026,15 +3966,17 @@ ${userMessage}`;
   }
   /**
    * Update cumulative token usage from API response
+   * @param usage - Token counts from the API response
+   * @param preCalculatedCost - Optional pre-calculated cost (from OpenRouter provider)
    */
-  updateTokenUsage(usage) {
+  updateTokenUsage(usage, preCalculatedCost) {
     const model = this.config.model ?? DEFAULT_MODEL;
     this.cumulativeTokenUsage.inputTokens = usage.input_tokens;
     this.cumulativeTokenUsage.outputTokens += usage.output_tokens;
     this.cumulativeTokenUsage.cacheCreationInputTokens = usage.cache_creation_input_tokens || 0;
     this.cumulativeTokenUsage.cacheReadInputTokens = usage.cache_read_input_tokens || 0;
     this.cumulativeTokenUsage.totalTokens = this.cumulativeTokenUsage.inputTokens + this.cumulativeTokenUsage.outputTokens;
-    const callCost = calculateCost(
+    const callCost = preCalculatedCost ?? calculateCost(
       model,
       usage.input_tokens,
       usage.output_tokens,
@@ -4135,16 +4077,13 @@ ${userMessage}`;
   /**
    * Compact the conversation history to reduce token usage.
    * 
-   * This uses Claude to create a structured summary of the conversation,
+   * This uses the current LLM to create a structured summary of the conversation,
    * then replaces the history with just the summary. This dramatically
    * reduces token count while preserving important context.
    * 
    * @returns Object with original/new token counts and the summary
    */
   async compactHistory() {
-    const model = this.config.model ?? "claude-sonnet-4-5-20250929";
-    const systemPrompt = this.config.systemPrompt ?? DEFAULT_SYSTEM_PROMPT;
-    const allTools = await this.getAllTools();
     if (this.conversationHistory.length < 2) {
       return {
         success: false,
@@ -4154,23 +4093,54 @@ ${userMessage}`;
       };
     }
     try {
-      const result = await compactConversation(
-        this.anthropic,
-        this.conversationHistory,
-        model,
-        systemPrompt,
-        allTools
-      );
-      this.conversationHistory = result.newHistory;
+      const originalContentLength = JSON.stringify(this.conversationHistory).length;
+      const originalTokens = Math.ceil(originalContentLength / 4);
+      const compactionPrompt = `Your context window is filling up. Create a concise summary of our conversation so far.
+
+Include:
+- User's main goals and what was accomplished
+- Files created/modified (with paths)
+- Key decisions and discoveries  
+- Next steps still needed
+- Any important context to preserve
+
+Be thorough but concise. The goal is to capture everything needed to continue seamlessly.`;
+      const compactionMessages = [
+        ...this.conversationHistory,
+        { role: "user", content: compactionPrompt }
+      ];
+      let summary;
+      if (this.config.provider === "openrouter" && this.llmProvider) {
+        const response = await this.llmProvider.chat(compactionMessages);
+        summary = response.text;
+      } else {
+        const model = this.config.model ?? DEFAULT_MODEL;
+        const response = await this.anthropic.messages.create({
+          model,
+          max_tokens: 4096,
+          messages: compactionMessages
+        });
+        const textBlocks = response.content.filter((block) => block.type === "text");
+        summary = textBlocks.map((block) => block.text).join("\n");
+      }
+      if (!summary || summary.trim().length === 0) {
+        throw new Error("Failed to generate summary");
+      }
+      const newHistory = [
+        { role: "assistant", content: summary.trim() }
+      ];
+      const newContentLength = JSON.stringify(newHistory).length;
+      const newTokens = Math.ceil(newContentLength / 4);
+      this.conversationHistory = newHistory;
       this.resetTokenUsage();
-      this.cumulativeTokenUsage.inputTokens = result.newTokens;
-      this.cumulativeTokenUsage.totalTokens = result.newTokens;
+      this.cumulativeTokenUsage.inputTokens = newTokens;
+      this.cumulativeTokenUsage.totalTokens = newTokens;
       this.config.onTokenUsage?.(this.cumulativeTokenUsage);
       return {
         success: true,
-        summary: result.summary,
-        originalTokenCount: result.originalTokens,
-        newTokenCount: result.newTokens
+        summary: summary.trim(),
+        originalTokenCount: originalTokens,
+        newTokenCount: newTokens
       };
     } catch (error2) {
       return {
@@ -5095,16 +5065,22 @@ Use /load <id> to load a session.`
       setIsProcessing(false);
       setThinkingText(null);
       abortController.current = null;
-      if (hasQueuedMessage && queuedInput) {
-        const nextMessage = queuedInput;
-        setQueuedInput("");
-        setHasQueuedMessage(false);
-        setTimeout(() => {
-          handleSubmit(nextMessage);
-        }, 100);
-      }
     }
-  }, [agent, isProcessing, isInterrupted, exit, onExit, handleSlashCommand, hasQueuedMessage, queuedInput]);
+  }, [agent, isProcessing, isInterrupted, exit, onExit, handleSlashCommand]);
+  useEffect(() => {
+    if (!isProcessing && hasQueuedMessage && queuedInput) {
+      const nextMessage = queuedInput;
+      setQueuedInput("");
+      setHasQueuedMessage(false);
+      setMessages((prev) => prev.filter(
+        (m) => !(m.role === "system" && m.content.startsWith("\u{1F4E5} Queued:"))
+      ));
+      const timer = setTimeout(() => {
+        handleSubmit(nextMessage);
+      }, 150);
+      return () => clearTimeout(timer);
+    }
+  }, [isProcessing, hasQueuedMessage, queuedInput, handleSubmit]);
   useEffect(() => {
     const originalConfig = agent.config;
     agent.config = {
@@ -5159,6 +5135,14 @@ Stopped ${count} background process${count > 1 ? "es" : ""}.`);
       onExit?.();
       exit();
     }
+    if (key.backspace && input === "" && hasQueuedMessage && queuedInput) {
+      setInput(queuedInput);
+      setQueuedInput("");
+      setHasQueuedMessage(false);
+      setMessages((prev) => prev.filter(
+        (m) => !(m.role === "system" && m.content.startsWith("\u{1F4E5} Queued:"))
+      ));
+    }
     if (key.escape && isProcessing) {
       setIsInterrupted(true);
       abortController.current?.abort();
@@ -5235,9 +5219,10 @@ Stopped ${count} background process${count > 1 ? "es" : ""}.`);
       "\u274C Error: ",
       error2
     ] }) }),
-    isProcessing && !streamingText && currentToolCalls.length === 0 && /* @__PURE__ */ jsx(Box, { marginBottom: 1, children: /* @__PURE__ */ jsxs(Text, { color: "cyan", children: [
+    isProcessing && /* @__PURE__ */ jsx(Box, { marginBottom: 1, children: /* @__PURE__ */ jsxs(Text, { color: "cyan", children: [
       /* @__PURE__ */ jsx(Spinner, { type: "dots" }),
-      " Thinking..."
+      " ",
+      currentToolCalls.length > 0 ? `Working... (${currentToolCalls.filter((tc) => tc.pending).length} tool${currentToolCalls.filter((tc) => tc.pending).length !== 1 ? "s" : ""} running)` : streamingText ? "Generating..." : "Thinking..."
     ] }) }),
     input.startsWith("/") && !isProcessing && /* @__PURE__ */ jsxs(Box, { flexDirection: "column", marginBottom: 1, paddingLeft: 2, children: [
       /* @__PURE__ */ jsx(Text, { color: "gray", dimColor: true, children: "Commands:" }),
@@ -5299,7 +5284,7 @@ Stopped ${count} background process${count > 1 ? "es" : ""}.`);
 }
 
 // src/index.ts
-var VERSION = "0.1.22";
+var VERSION = "0.1.24";
 function cleanup() {
   if (processManager.hasRunning()) {
     const count = processManager.runningCount();

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@quantish/agent",
-  "version": "0.1.22",
+  "version": "0.1.24",
   "description": "AI-powered agent for building trading bots on Polymarket",
   "type": "module",
   "bin": {