drivecore · bhouston · Mar 21, 2025 · Mar 21, 2025 · Mar 21, 2025 · Mar 21, 2025
diff --git a/README.md b/README.md
@@ -12,6 +12,7 @@ Command-line interface for AI-powered coding tasks. Full details available on th
 - 👤 **Human Compatible**: Uses README.md, project files and shell commands to build its own context
 - 🌐 **GitHub Integration**: GitHub mode for working with issues and PRs as part of workflow
 - 📄 **Model Context Protocol**: Support for MCP to access external context sources
+- 🧠 **Message Compaction**: Automatic management of context window for long-running agents
 
 Please join the MyCoder.ai discord for support: https://discord.gg/5K6TYrHGHt
 

diff --git a/docs/features/message-compaction.md b/docs/features/message-compaction.md
@@ -0,0 +1,105 @@
+# Message Compaction
+
+When agents run for extended periods, they accumulate a large history of messages that eventually fills up the LLM's context window, causing errors when the token limit is exceeded. The message compaction feature helps prevent this by providing agents with awareness of their token usage and tools to manage their context window.
+
+## Features
+
+### 1. Token Usage Tracking
+
+The LLM abstraction now tracks and returns:
+- Total tokens used in the current completion request
+- Maximum allowed tokens for the model/provider
+
+This information is used to monitor context window usage and trigger appropriate actions.
+
+### 2. Status Updates
+
+Agents receive status updates with information about:
+- Current token usage and percentage of the maximum
+- Cost so far
+- Active sub-agents and their status
+- Active shell processes and their status
+- Active browser sessions and their status
+
+Status updates are sent:
+1. Every 5 agent interactions (periodic updates)
+2. Whenever token usage exceeds 50% of the maximum (threshold-based updates)
+
+Example status update:
+```
+--- STATUS UPDATE ---
+Token Usage: 45,235/100,000 (45%)
+Cost So Far: $0.23
+
+Active Sub-Agents: 2
+- sa_12345: Analyzing project structure and dependencies
+- sa_67890: Implementing unit tests for compactHistory tool
+
+Active Shell Processes: 3
+- sh_abcde: npm test
+- sh_fghij: npm run watch
+- sh_klmno: git status
+
+Active Browser Sessions: 1
+- bs_12345: https://www.typescriptlang.org/docs/handbook/utility-types.html
+
+If token usage is high (>70%), consider using the 'compactHistory' tool to reduce context size.
+--- END STATUS ---
+```
+
+### 3. Message Compaction Tool
+
+The `compactHistory` tool allows agents to compact their message history by summarizing older messages while preserving recent context. This tool:
+
+1. Takes a parameter for how many recent messages to preserve unchanged
+2. Summarizes all older messages into a single, concise summary
+3. Replaces the original messages with the summary and preserved messages
+4. Reports on the reduction in context size
+
+## Usage
+
+Agents are instructed to monitor their token usage through status updates and use the `compactHistory` tool when token usage approaches 50% of the maximum:
+
+```javascript
+// Example of agent using the compactHistory tool
+{
+  name: "compactHistory",
+  preserveRecentMessages: 10,
+  customPrompt: "Focus on summarizing our key decisions and current tasks."
+}
+```
+
+## Configuration
+
+The message compaction feature is enabled by default with reasonable defaults:
+- Status updates every 5 agent interactions
+- Recommendation to compact at 70% token usage
+- Default preservation of 10 recent messages when compacting
+
+## Model Token Limits
+
+The system includes token limits for various models:
+
+### Anthropic Models
+- claude-3-opus-20240229: 200,000 tokens
+- claude-3-sonnet-20240229: 200,000 tokens
+- claude-3-haiku-20240307: 200,000 tokens
+- claude-2.1: 100,000 tokens
+
+### OpenAI Models
+- gpt-4o: 128,000 tokens
+- gpt-4-turbo: 128,000 tokens
+- gpt-3.5-turbo: 16,385 tokens
+
+### Ollama Models
+- llama2: 4,096 tokens
+- mistral: 8,192 tokens
+- mixtral: 32,768 tokens
+
+## Benefits
+
+- Prevents context window overflow errors
+- Maintains important context for agent operation
+- Enables longer-running agent sessions
+- Makes the system more robust for complex tasks
+- Gives agents self-awareness of resource usage
diff --git a/example-status-update.md b/example-status-update.md
@@ -0,0 +1,50 @@
+# Example Status Update
+
+This is an example of what the status update looks like for the agent:
+
+```
+--- STATUS UPDATE ---
+Token Usage: 45,235/100,000 (45%)
+Cost So Far: $0.23
+
+Active Sub-Agents: 2
+- sa_12345: Analyzing project structure and dependencies
+- sa_67890: Implementing unit tests for compactHistory tool
+
+Active Shell Processes: 3
+- sh_abcde: npm test -- --watch packages/agent/src/tools/utility
+- sh_fghij: npm run watch
+- sh_klmno: git status
+
+Active Browser Sessions: 1
+- bs_12345: https://www.typescriptlang.org/docs/handbook/utility-types.html
+
+Your token usage is high (45%). It is recommended to use the 'compactHistory' tool now to reduce context size.
+--- END STATUS ---
+```
+
+## About Status Updates
+
+Status updates are sent to the agent (every 5 interactions and whenever token usage exceeds 50%) to provide awareness of:
+
+1. **Token Usage**: Current usage and percentage of maximum context window
+2. **Cost**: Estimated cost of the session so far
+3. **Active Sub-Agents**: Running background agents and their tasks
+4. **Active Shell Processes**: Running shell commands
+5. **Active Browser Sessions**: Open browser sessions and their URLs
+
+When token usage gets high (>70%), the agent is reminded to use the `compactHistory` tool to reduce context size by summarizing older messages.
+
+## Using the compactHistory Tool
+
+The agent can use the compactHistory tool like this:
+
+```javascript
+{
+  name: "compactHistory",
+  preserveRecentMessages: 10,
+  customPrompt: "Optional custom summarization prompt"
+}
+```
+
+This will summarize all but the 10 most recent messages into a single summary message, significantly reducing token usage while preserving important context.
diff --git a/packages/agent/CHANGELOG.md b/packages/agent/CHANGELOG.md
@@ -1,3 +1,10 @@
+# [mycoder-agent-v1.6.0](https://github.com/drivecore/mycoder/compare/mycoder-agent-v1.5.0...mycoder-agent-v1.6.0) (2025-03-21)
+
+
+### Features
+
+* **browser:** add system browser detection for Playwright ([00bd879](https://github.com/drivecore/mycoder/commit/00bd879443c9de51c6ee5e227d4838905506382a)), closes [#333](https://github.com/drivecore/mycoder/issues/333)
+
 # [mycoder-agent-v1.5.0](https://github.com/drivecore/mycoder/compare/mycoder-agent-v1.4.2...mycoder-agent-v1.5.0) (2025-03-20)
 
 ### Bug Fixes

diff --git a/packages/agent/package.json b/packages/agent/package.json
@@ -1,6 +1,6 @@
 {
   "name": "mycoder-agent",
-  "version": "1.5.0",
+  "version": "1.6.0",
   "description": "Agent module for mycoder - an AI-powered software development assistant",
   "type": "module",
   "main": "dist/index.js",

diff --git a/packages/agent/src/core/llm/providers/anthropic.ts b/packages/agent/src/core/llm/providers/anthropic.ts
@@ -81,13 +81,33 @@ function addCacheControlToMessages(
   });
 }
 
-function tokenUsageFromMessage(message: Anthropic.Message) {
+// Define model context window sizes for Anthropic models
+const ANTHROPIC_MODEL_LIMITS: Record<string, number> = {
+  'claude-3-opus-20240229': 200000,
+  'claude-3-sonnet-20240229': 200000,
+  'claude-3-haiku-20240307': 200000,
+  'claude-3-7-sonnet-20250219': 200000,
+  'claude-2.1': 100000,
+  'claude-2.0': 100000,
+  'claude-instant-1.2': 100000,
+  // Add other models as needed
+};
+
+function tokenUsageFromMessage(message: Anthropic.Message, model: string) {
   const usage = new TokenUsage();
   usage.input = message.usage.input_tokens;
   usage.cacheWrites = message.usage.cache_creation_input_tokens ?? 0;
   usage.cacheReads = message.usage.cache_read_input_tokens ?? 0;
   usage.output = message.usage.output_tokens;
-  return usage;
+
+  const totalTokens = usage.input + usage.output;
+  const maxTokens = ANTHROPIC_MODEL_LIMITS[model] || 100000; // Default fallback
+
+  return {
+    usage,
+    totalTokens,
+    maxTokens,
+  };
 }
 
 /**
@@ -175,10 +195,14 @@ export class AnthropicProvider implements LLMProvider {
           };
         });
 
+      const tokenInfo = tokenUsageFromMessage(response, this.model);
+
       return {
         text: content,
         toolCalls: toolCalls,
-        tokenUsage: tokenUsageFromMessage(response),
+        tokenUsage: tokenInfo.usage,
+        totalTokens: tokenInfo.totalTokens,
+        maxTokens: tokenInfo.maxTokens,
       };
     } catch (error) {
       throw new Error(

diff --git a/packages/agent/src/core/llm/providers/ollama.ts b/packages/agent/src/core/llm/providers/ollama.ts
@@ -13,6 +13,22 @@ import {
 
 import { TokenUsage } from '../../tokens.js';
 import { ToolCall } from '../../types.js';
+// Define model context window sizes for Ollama models
+// These are approximate and may vary based on specific model configurations
+const OLLAMA_MODEL_LIMITS: Record<string, number> = {
+  'llama2': 4096,
+  'llama2-uncensored': 4096,
+  'llama2:13b': 4096,
+  'llama2:70b': 4096,
+  'mistral': 8192,
+  'mistral:7b': 8192,
+  'mixtral': 32768,
+  'codellama': 16384,
+  'phi': 2048,
+  'phi2': 2048,
+  'openchat': 8192,
+  // Add other models as needed
+};
 import { LLMProvider } from '../provider.js';
 import {
   GenerateOptions,
@@ -56,7 +72,7 @@ export class OllamaProvider implements LLMProvider {
       messages,
       functions,
       temperature = 0.7,
-      maxTokens,
+      maxTokens: requestMaxTokens,
       topP,
       frequencyPenalty,
       presencePenalty,
@@ -86,10 +102,10 @@ export class OllamaProvider implements LLMProvider {
     };
 
     // Add max_tokens if provided
-    if (maxTokens !== undefined) {
+    if (requestMaxTokens !== undefined) {
       requestOptions.options = {
         ...requestOptions.options,
-        num_predict: maxTokens,
+        num_predict: requestMaxTokens,
       };
     }
 
@@ -114,11 +130,23 @@ export class OllamaProvider implements LLMProvider {
     const tokenUsage = new TokenUsage();
     tokenUsage.output = response.eval_count || 0;
     tokenUsage.input = response.prompt_eval_count || 0;
+
+    // Calculate total tokens and get max tokens for the model
+    const totalTokens = tokenUsage.input + tokenUsage.output;
+
+    // Extract the base model name without specific parameters
+    const baseModelName = this.model.split(':')[0];
+    // Check if model exists in limits, otherwise use base model or default
+    const modelMaxTokens = OLLAMA_MODEL_LIMITS[this.model] || 
+                          (baseModelName ? OLLAMA_MODEL_LIMITS[baseModelName] : undefined) || 
+                          4096; // Default fallback
 
     return {
       text: content,
       toolCalls: toolCalls,
       tokenUsage: tokenUsage,
+      totalTokens,
+      maxTokens: modelMaxTokens,
     };
   }
 

diff --git a/packages/agent/src/core/llm/providers/openai.ts b/packages/agent/src/core/llm/providers/openai.ts
@@ -4,7 +4,7 @@
 import OpenAI from 'openai';
 
 import { TokenUsage } from '../../tokens.js';
-import { ToolCall } from '../../types';
+import { ToolCall } from '../../types.js';
 import { LLMProvider } from '../provider.js';
 import {
   GenerateOptions,
@@ -19,6 +19,19 @@ import type {
   ChatCompletionTool,
 } from 'openai/resources/chat';
 
+// Define model context window sizes for OpenAI models
+const OPENAI_MODEL_LIMITS: Record<string, number> = {
+  'gpt-4o': 128000,
+  'gpt-4-turbo': 128000,
+  'gpt-4-0125-preview': 128000,
+  'gpt-4-1106-preview': 128000,
+  'gpt-4': 8192,
+  'gpt-4-32k': 32768,
+  'gpt-3.5-turbo': 16385,
+  'gpt-3.5-turbo-16k': 16385,
+  // Add other models as needed
+};
+
 /**
  * OpenAI-specific options
  */
@@ -60,7 +73,7 @@ export class OpenAIProvider implements LLMProvider {
       messages,
       functions,
       temperature = 0.7,
-      maxTokens,
+      maxTokens: requestMaxTokens,
       stopSequences,
       topP,
       presencePenalty,
@@ -79,7 +92,7 @@ export class OpenAIProvider implements LLMProvider {
         model: this.model,
         messages: formattedMessages,
         temperature,
-        max_tokens: maxTokens,
+        max_tokens: requestMaxTokens,
         stop: stopSequences,
         top_p: topP,
         presence_penalty: presencePenalty,
@@ -116,11 +129,17 @@ export class OpenAIProvider implements LLMProvider {
       const tokenUsage = new TokenUsage();
       tokenUsage.input = response.usage?.prompt_tokens || 0;
       tokenUsage.output = response.usage?.completion_tokens || 0;
+
+      // Calculate total tokens and get max tokens for the model
+      const totalTokens = tokenUsage.input + tokenUsage.output;
+      const modelMaxTokens = OPENAI_MODEL_LIMITS[this.model] || 8192; // Default fallback
 
       return {
         text: content,
         toolCalls,
         tokenUsage,
+        totalTokens,
+        maxTokens: modelMaxTokens,
       };
     } catch (error) {
       throw new Error(`Error calling OpenAI API: ${(error as Error).message}`);
@@ -198,4 +217,4 @@ export class OpenAIProvider implements LLMProvider {
       },
     }));
   }
-}
+}
diff --git a/packages/agent/src/core/llm/types.ts b/packages/agent/src/core/llm/types.ts
@@ -80,6 +80,9 @@
   text: string;
   toolCalls: ToolCall[];
   tokenUsage: TokenUsage;
+  // Add new fields for context window tracking
+  totalTokens?: number;  // Total tokens used in this request
+  maxTokens?: number;    // Maximum allowed tokens for this model
 }
 
 /**
@@ -104,5 +107,5 @@
  apiKey?: string;
  baseUrl?: string;
  organization?: string;
  [key: string]: any; // Allow for provider-specific options
 }