add token caching, remove mention of other LLM providers no longer supported.

Author: bhouston

docs/getting-started/index.mdx

@@ -41,10 +41,6 @@ Before using MyCoder with a specific provider, you need to provide the appropria
    export ANTHROPIC_API_KEY=your-api-key
    # or
    export OPENAI_API_KEY=your-api-key
-   # or
-   export MISTRAL_API_KEY=your-api-key
-   # or
-   export XAI_API_KEY=your-api-key
    ```
 
 2. Create a `.env` file in your working directory with the appropriate key:
@@ -61,10 +57,8 @@ MyCoder supports multiple AI providers:
 | Provider   | Environment Variable | Models                               |
 | ---------- | -------------------- | ------------------------------------ |
 | Anthropic  | `ANTHROPIC_API_KEY`  | claude-3-opus, claude-3-sonnet, etc. |
-| OpenAI     | `OPENAI_API_KEY`     | gpt-4o, o3-mini, etc.                |
-| Mistral AI | `MISTRAL_API_KEY`    | mistral-large, mistral-medium, etc.  |
-| xAI/Grok   | `XAI_API_KEY`        | grok-1                               |
-| Ollama     | N/A (local)          | Various local models                 |
+| OpenAI     | `OPENAI_API_KEY`     | gpt-4o, gpt-4-turbo, etc.            |
+| Ollama     | N/A (local)          | Models with tool calling support     |
 
 You can specify which provider and model to use with the `--provider` and `--model` options:
 

docs/index.md

@@ -14,7 +14,7 @@ Currently available as a research preview, MyCoder is built to work alongside de
 
 ## Key Features
 
-- **AI-Powered**: Supports multiple AI providers including Anthropic, OpenAI, Mistral AI, xAI/Grok, and Ollama
+- **AI-Powered**: Supports multiple AI providers including Anthropic, OpenAI, and Ollama
 - **Extensible Tool System**: Includes tools for file operations, shell commands, web browsing, and more
 - **Parallel Execution**: Can spawn sub-agents to work on different parts of a task simultaneously
 - **Self-Modification**: Capable of modifying code, including its own codebase

docs/providers/anthropic.md

@@ -58,6 +58,25 @@ Anthropic offers several Claude models with different capabilities and price poi
 - Claude models have a 200K token context window, allowing for large codebases to be processed
 - For cost-sensitive applications, consider using Claude Haiku for simpler tasks
 
+## Token Caching
+
+MyCoder implements token caching for Anthropic's Claude models to optimize performance and reduce API costs:
+
+- Token caching stores and reuses parts of the conversation history
+- The Anthropic provider uses Claude's native cache control mechanisms
+- This significantly reduces token usage for repeated or similar queries
+- Cache efficiency is automatically optimized based on conversation context
+
+You can enable or disable token caching in your configuration:
+
+```javascript
+export default {
+  provider: 'anthropic',
+  model: 'claude-3-7-sonnet-20250219',
+  tokenCache: true, // Enable token caching (default is true)
+};
+```
+
 ## Troubleshooting
 
 If you encounter issues with Anthropic's Claude:

docs/providers/openai.md

@@ -51,13 +51,15 @@ export default {
 
 ## Supported Models
 
-OpenAI offers several models with different capabilities:
+MyCoder supports all OpenAI models that have tool/function calling capabilities. Here are some recommended models:
 
 - `gpt-4o` (recommended) - Latest model with strong reasoning and tool-calling capabilities
 - `gpt-4-turbo` - Strong performance with 128K context window
 - `gpt-4` - Original GPT-4 model with 8K context window
 - `gpt-3.5-turbo` - More affordable option for simpler tasks
 
+You can use any other OpenAI model that supports function calling with MyCoder. The OpenAI provider is not limited to just these listed models.
+
 ## Best Practices
 
 - GPT-4o provides the best balance of performance and cost for most MyCoder tasks