docs: Update with model download instructions.

Author: daavoo

README.md

@@ -2,10 +2,21 @@
 
 Building blocks for **local** agents in C++.
 
-> **Note:** This library is designed for running small language models locally using [llama.cpp](https://github.com/ggml-org/llama.cpp). It does not support cloud-based APIs (OpenAI, Anthropic, etc.). If you want to call external LLM APIs, this is not the right fit.
+> [!NOTE]
+> This library is designed for running small language models locally using [llama.cpp](https://github.com/ggml-org/llama.cpp). If you want to call external LLM APIs, this is not the right fit.
 
 ## Examples
 
+You need to download a GGUF model in order to run the examples, the default model configuration is set for `granite-4.0-micro`:
+
+```bash
+wget https://huggingface.co/ibm-granite/granite-4.0-micro-GGUF/resolve/main/granite-4.0-micro-Q8_0.gguf
+```
+
+> [!IMPORTANT]
+> If you use a different model, you will probably have to adjust the values in `ModelConfig`.
+
+
 - **[Context Engineering](./examples/context-engineering/README.md)** - Use callbacks to manipulate the context between iterations of the agent loop.
 
 - **[Memory](./examples/memory/README.md)** - Use tools that allow an agent to store and retrieve relevant information across conversations.

examples/context-engineering/README.md

@@ -4,7 +4,15 @@ This example demonstrates how to use callbacks to modify the input messages (con
 
 ## Building Blocks
 
+### Callbacks
 
+This example uses three callbacks:
+
+- **ContextTrimmerCallback**: Implements `before_llm_call` to modify messages before they are sent to the LLM. It keeps only the N most recent tool call pairs (assistant message with tool_calls + tool responses), trimming older ones to prevent context window overflow during long conversations.
+
+- **LoggingCallback**: Shared callback from `examples/shared/` that logs tool execution information, displaying which tool is being called and its results.
+
+- **ErrorRecoveryCallback**: Shared callback from `examples/shared/` that converts tool execution errors into JSON results, allowing the agent to see errors and potentially retry or adjust.
 
 ### Tools
 

examples/shell/README.md

@@ -12,6 +12,11 @@ This example exposes a single tool that can execute the given shell commands:
 
 ### Callbacks
 
+This example uses two callbacks:
+
+- **ShellConfirmationCallback**: Implements human-in-the-loop confirmation before executing shell commands. Before each command runs, it prompts the user to approve, reject, or edit the command. This is critical for security when executing arbitrary shell commands.
+
+- **ErrorRecoveryCallback**: Shared callback from `examples/shared/` that converts tool execution errors into JSON results, allowing the agent to see errors and potentially retry or adjust.
 
 ## Building
 
@@ -36,12 +41,11 @@ cmake --build build -j$(nproc)
 ## Usage
 
 ```bash
-./shell-example -m <path-to-model.gguf> [-d <working-directory>]
+./shell-example -m <path-to-model.gguf>
 ```
 
 Options:
 - `-m <path>` - Path to the GGUF model file (required)
-- `-d <path>` - Working directory for shell commands (default: current directory)
 
 ## Example