diff --git a/.prompts/docs-review.md b/.prompts/docs-review.md
index 42f67915e..6917586b7 100644
--- a/.prompts/docs-review.md
+++ b/.prompts/docs-review.md
@@ -5,8 +5,18 @@ Review documentation for accuracy, completeness, and consistency. Focus on thing
 Don't waste time on these—CI and pre-commit hooks handle them:
 
 - **README help output**: `markdown-code-runner` regenerates `agent-cli --help` blocks
+- **Options tables**: `docs_gen` module auto-generates options from CLI introspection
 - **Linting/formatting**: Handled by pre-commit
 
+The `docs_gen` module (`agent_cli/docs_gen.py`) provides:
+- `all_options_for_docs(cmd)`: Complete options tables grouped by panel
+- `env_vars_table()`: Environment variables documentation
+- `provider_matrix()`: Provider comparison table
+- `config_example(cmd)`: Example TOML configuration
+- `commands_table()`: Commands overview table
+
+Run `uv run python docs/update_docs.py` to regenerate all auto-generated content.
+
 ## What This Review Is For
 
 Focus on things that require judgment:
@@ -31,53 +41,41 @@ git diff --name-only HEAD~20 | grep "\.py$"
 
 Look for new features, changed defaults, renamed options, or removed functionality.
 
-### 2. Verify Command Documentation (HIGH PRIORITY)
+### 2. Verify Command Documentation
 
-**Options tables in `docs/commands/` are manually maintained and frequently drift from the actual CLI.** This is the most common source of documentation errors.
+Options tables are now auto-generated, so focus on what's NOT automated:
 
-For EACH command doc file, systematically compare against the actual CLI:
+**Check for missing command docs:**
 
 ```bash
-# Get list of all commands
+# Compare commands in CLI vs docs that exist
 agent-cli --help
-
-# For each command, dump actual options and compare to docs
-agent-cli transcribe --help
-agent-cli chat --help
-agent-cli autocorrect --help
-agent-cli speak --help
-agent-cli voice-edit --help
-agent-cli assistant --help
-agent-cli transcribe-daemon --help
-agent-cli rag-proxy --help
-agent-cli memory --help
-agent-cli server --help
+ls docs/commands/*.md
 ```
 
-**Check each option in the help output against the docs:**
+Every command should have a corresponding `docs/commands/<command>.md` file. When adding a new command doc, also update:
+- `docs/commands/index.md` - add to the commands table
+- `zensical.toml` - add to the `nav` sidebar under Commands
 
-| Check | Common Issues |
-|-------|---------------|
-| Option exists in docs? | New options added to CLI but not documented |
-| Option still exists in CLI? | Removed options still in docs |
-| Default value correct? | Defaults change, docs not updated |
-| Short flag correct? | `-m` vs `-M`, missing short flags |
-| Description accurate? | Behavior changed, description stale |
-| Type correct? | `PATH` vs `TEXT`, `INTEGER` vs `FLOAT` |
+**What still needs manual review:**
 
-**Also check `agent_cli/opts.py`** - this defines shared options used across commands. Changes here affect multiple commands.
+| Check | What to Look For |
+|-------|------------------|
+| Description accuracy | Does the prose description match actual behavior? |
+| Example commands | Would these actually work? Are they useful? |
+| Workflow explanations | Is the step-by-step flow still accurate? |
+| Use cases | Are suggested use cases realistic? |
+| Cross-links | Do links to related commands work? |
 
-**Check for missing command docs:**
+**Verify auto-generation is working:**
 
 ```bash
-# Compare commands in CLI vs docs that exist
-agent-cli --help
-ls docs/commands/*.md
+# Run update script and check for changes
+uv run python docs/update_docs.py
+git diff docs/
 ```
 
-Every command should have a corresponding `docs/commands/<command>.md` file. When adding a new command doc, also update:
-- `docs/commands/index.md` - add to the commands table
-- `zensical.toml` - add to the `nav` sidebar under Commands
+If the script produces changes, either commit them or investigate why docs drifted.
 
 ### 3. Verify docs/configuration.md
 
@@ -118,19 +116,19 @@ Check:
 
 These are particularly prone to errors when docs are AI-generated or AI-maintained:
 
-| Area | How to Verify | Past Issues |
-|------|---------------|-------------|
-| **Model names** | Check `agent_cli/opts.py` defaults | `gpt-4o-mini` → `gpt-5-mini` |
-| **Tool/function names** | Check `agent_cli/_tools.py` | `web_search` didn't exist, was `duckduckgo_search` |
-| **File paths** | Grep for `PID_DIR`, `CONFIG_DIR`, etc. | PID files documented in wrong directory |
+| Area | How to Verify | Notes |
+|------|---------------|-------|
+| **Tool/function names** | Check `agent_cli/_tools.py` | Tool names in prose/examples may drift |
+| **File paths** | Grep for `PID_DIR`, `CONFIG_DIR`, etc. | Paths in prose may be wrong |
 | **Dependencies** | Compare against `pyproject.toml` | Listed packages that don't exist |
-| **Environment variables** | Check `envvar=` in opts.py | Undocumented or renamed env vars |
 | **Provider names** | Check `agent_cli/services/` | Providers listed that aren't implemented |
 
-```bash
-# Verify model defaults
-grep -E "model.*=.*\"" agent_cli/opts.py
+**Now auto-generated (lower risk):**
+- Model defaults → captured in options tables via `docs_gen`
+- Environment variables → use `env_vars_table()` for accuracy
+- Option defaults/types → auto-generated from CLI introspection
 
+```bash
 # Verify tool names
 grep -E "def.*_tool|Tool\(" agent_cli/_tools.py
 
@@ -140,8 +138,8 @@ grep -rE "(PID_DIR|CONFIG_DIR|CACHE_DIR)" agent_cli/
 # Verify dependencies
 cat pyproject.toml | grep -A 50 "dependencies"
 
-# Verify env vars
-grep "envvar=" agent_cli/opts.py
+# Verify providers match implementations
+ls agent_cli/services/
 ```
 
 ### 6. Check Examples
@@ -155,10 +153,12 @@ For examples in any doc:
 
 The same info appears in multiple places. Check for conflicts:
 - README.md vs docs/index.md
-- docs/commands/*.md vs actual CLI help
+- Prose/examples in docs vs actual CLI behavior
 - docs/configuration.md vs agent_cli/example-config.toml
 - Provider/port info across architecture docs
 
+Note: Options tables are auto-generated, so conflicts there indicate the update script wasn't run.
+
 ### 8. Cross-Links for Navigation
 
 When commands are mentioned in prose or examples, they should link to their documentation pages. This improves discoverability and user navigation.
@@ -215,8 +215,8 @@ For each issue, provide a ready-to-apply fix:
 ```
 ### Issue: [Brief description]
 
-- **File**: docs/commands/chat.md:45
-- **Problem**: `--history-dir` default shown as `~/.chat-history` but actual default is `~/.config/agent-cli/history`
-- **Fix**: Update the default value in the options table
-- **Verify**: `agent-cli chat --help`
+- **File**: docs/commands/chat.md:25
+- **Problem**: Example uses `--model gpt-4` but the default model is now `gpt-4o-mini`
+- **Fix**: Update the example to use the current default or a valid model name
+- **Verify**: `agent-cli chat --help` shows current default
 ```
diff --git a/CLAUDE.md b/CLAUDE.md
new file mode 100644
index 000000000..68ad951ca
--- /dev/null
+++ b/CLAUDE.md
@@ -0,0 +1,114 @@
+# CLAUDE.md
+
+This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.
+
+## Development Rules
+
+- Prefer functional style Python over classes with inheritance
+- Keep it DRY - reuse code, check for existing patterns before adding new ones
+- Implement the simplest solution; don't generalize until needed
+- Only implement what's asked for, nothing extra
+- Always run `pytest` before claiming a task is done
+- Run `pre-commit run --all-files` before committing
+- Use `git add <specific-file>` not `git add .`
+- CLI help in README.md is auto-generated - don't edit manually
+
+## Build & Development Commands
+
+```bash
+# Install all dependencies (recommended for development)
+uv sync --all-extras
+
+# Or install specific extras only
+uv sync --extra rag      # RAG proxy dependencies
+uv sync --extra memory   # Memory proxy dependencies
+uv sync --extra vad      # Voice activity detection (transcribe-daemon)
+
+# Run the CLI during development
+uv run agent-cli <command>
+
+# Run tests (coverage enabled by default via pyproject.toml)
+uv run pytest
+
+# Linting (pre-commit runs ruff + other checks)
+pre-commit run --all-files
+
+# Update auto-generated documentation (CODE:START blocks in markdown)
+uv run python docs/update_docs.py
+```
+
+## Architecture Overview
+
+### CLI Structure
+
+The CLI is built with **Typer**. Entry point is `agent_cli/cli.py` which registers all commands.
+
+**Shared Options Pattern**: Common CLI options (providers, API keys, audio devices) are defined once in `agent_cli/opts.py` and reused across commands. This ensures consistency and enables auto-generated documentation.
+
+### Provider Abstraction
+
+The codebase uses a **provider pattern** for AI services, allowing switching between local and cloud backends:
+
+| Capability | Providers | Implementation |
+|------------|-----------|----------------|
+| ASR (Speech-to-Text) | `wyoming`, `openai` | `services/asr.py` |
+| LLM | `ollama`, `openai`, `gemini` | `services/llm.py` |
+| TTS (Text-to-Speech) | `wyoming`, `openai`, `kokoro` | `services/tts.py` |
+
+Each agent accepts `--{asr,llm,tts}-provider` flags to select the backend.
+
+### Key Modules
+
+```
+agent_cli/
+├── cli.py              # Typer app, command registration
+├── opts.py             # Shared CLI option definitions (single source of truth)
+├── config.py           # Config file loading, dataclasses for typed configs
+├── agents/             # CLI commands (one file per command)
+│   ├── transcribe.py   # Voice-to-text with optional LLM cleanup
+│   ├── autocorrect.py  # Grammar/spelling correction
+│   ├── chat.py         # Conversational agent with tools
+│   ├── voice_edit.py   # Voice commands on clipboard text
+│   └── ...
+├── services/           # Provider implementations
+│   ├── asr.py          # Wyoming/OpenAI transcription
+│   ├── llm.py          # Ollama/OpenAI/Gemini LLM calls
+│   └── tts.py          # Wyoming/OpenAI/Kokoro TTS
+├── core/               # Shared utilities
+│   ├── audio.py        # Audio recording, device selection
+│   ├── process.py      # Background process management (--toggle, --stop)
+│   └── utils.py        # Console output, logging setup
+├── rag/                # RAG proxy server implementation
+├── memory/             # Long-term memory proxy server
+└── docs_gen.py         # Auto-generates docs from CLI introspection
+```
+
+### Agent Pattern
+
+Each agent in `agents/` follows a consistent pattern:
+1. Import shared options from `opts.py`
+2. Define a Typer command decorated with `@app.command()`
+3. Use `config.py` dataclasses to group related options
+4. Call provider services from `services/`
+
+### Background Process Management
+
+Commands like `transcribe`, `voice-edit`, and `chat` support running as background processes with hotkey integration:
+- `--toggle`: Start if stopped, stop if running
+- `--stop`: Stop any running instance
+- `--status`: Check if running
+
+PID files are stored in `~/.cache/agent-cli/`.
+
+### Documentation Auto-Generation
+
+The `docs_gen` module introspects Typer commands to generate Markdown tables. Documentation files use `markdown-code-runner` markers:
+
+```markdown
+<!-- CODE:START -->
+<!-- from agent_cli.docs_gen import all_options_for_docs -->
+<!-- print(all_options_for_docs("transcribe")) -->
+<!-- CODE:END -->
+```
+
+Run `uv run python docs/update_docs.py` to regenerate all auto-generated content.
diff --git a/docs/architecture/index.md b/docs/architecture/index.md
index fca6c40c9..4bda9816e 100644
--- a/docs/architecture/index.md
+++ b/docs/architecture/index.md
@@ -44,7 +44,7 @@ Each AI capability (ASR, LLM, TTS) has multiple backend providers:
 
 | Provider | Implementation | GPU Support | Latency |
 |----------|---------------|-------------|---------|
-| `wyoming` | Wyoming Faster Whisper | CUDA/Metal | Low |
+| `wyoming` | Wyoming Whisper (faster-whisper/MLX) | CUDA/Metal | Low |
 | `openai` | OpenAI Whisper API | Cloud | Medium |
 
 ### LLM (Large Language Model)
diff --git a/docs/commands/install-hotkeys.md b/docs/commands/install-hotkeys.md
index da0c90137..b34dc6e18 100644
--- a/docs/commands/install-hotkeys.md
+++ b/docs/commands/install-hotkeys.md
@@ -16,18 +16,19 @@ agent-cli install-hotkeys [OPTIONS]
 
 Sets up hotkeys for common workflows:
 
-macOS:
+**macOS:**
+
 - Cmd+Shift+R: Toggle voice transcription
 - Cmd+Shift+A: Autocorrect clipboard text
 - Cmd+Shift+V: Voice edit clipboard text
 
-Linux:
+**Linux:**
+
 - Super+Shift+R: Toggle voice transcription
 - Super+Shift+A: Autocorrect clipboard text
 - Super+Shift+V: Voice edit clipboard text
 
-On macOS, you may need to grant Accessibility permissions to skhd in
-System Settings -> Privacy & Security -> Accessibility.
+On macOS, you may need to grant Accessibility permissions to skhd in System Settings → Privacy & Security → Accessibility.
 
 ## Options
 
diff --git a/docs/commands/install-services.md b/docs/commands/install-services.md
index eafb04fb9..1423132a5 100644
--- a/docs/commands/install-services.md
+++ b/docs/commands/install-services.md
@@ -17,7 +17,7 @@ agent-cli install-services [OPTIONS]
 Installs the following services (based on your OS):
 
 - Ollama (local LLM server)
-- Wyoming Faster Whisper (speech-to-text)
+- Wyoming Whisper (faster-whisper on Linux/Intel, MLX Whisper on Apple Silicon)
 - Wyoming Piper (text-to-speech)
 - Wyoming OpenWakeWord (wake word detection)
 
diff --git a/docs/commands/rag-proxy.md b/docs/commands/rag-proxy.md
index e452dc0b8..0fdaa0412 100644
--- a/docs/commands/rag-proxy.md
+++ b/docs/commands/rag-proxy.md
@@ -143,13 +143,13 @@ Any OpenAI-compatible client can use the RAG proxy:
 # curl
 curl http://localhost:8000/v1/chat/completions \
   -H "Content-Type: application/json" \
-  -d '{"model": "gpt-4", "messages": [{"role": "user", "content": "What do my notes say about X?"}]}'
+  -d '{"model": "<your-model>", "messages": [{"role": "user", "content": "What do my notes say about X?"}]}'
 
 # Python (openai library)
 from openai import OpenAI
 client = OpenAI(base_url="http://localhost:8000/v1", api_key="not-needed")
 response = client.chat.completions.create(
-    model="gpt-5-mini",
+    model="<your-model>",
     messages=[{"role": "user", "content": "Summarize my project notes"}]
 )
 ```
diff --git a/docs/commands/start-services.md b/docs/commands/start-services.md
index 3016e16cc..721720b23 100644
--- a/docs/commands/start-services.md
+++ b/docs/commands/start-services.md
@@ -17,7 +17,7 @@ agent-cli start-services [OPTIONS]
 Starts:
 
 - Ollama (LLM server)
-- Wyoming Faster Whisper (speech-to-text)
+- Wyoming Whisper (faster-whisper on Linux/Intel, MLX Whisper on Apple Silicon)
 - Wyoming Piper (text-to-speech)
 - Wyoming OpenWakeWord (wake word detection)
 
diff --git a/docs/commands/transcribe.md b/docs/commands/transcribe.md
index c79f8aa74..9d5d170cf 100644
--- a/docs/commands/transcribe.md
+++ b/docs/commands/transcribe.md
@@ -18,7 +18,7 @@ This command:
 
 1. Starts listening to your microphone immediately
 2. Records your speech
-3. When you press `Ctrl+C`, sends audio to a Whisper server
+3. When you press `Ctrl+C`, stops recording and finalizes transcription (Wyoming streams live; OpenAI uploads after stop)
 4. Copies the transcribed text to your clipboard
 5. Optionally uses an LLM to clean up the transcript
 
diff --git a/docs/index.md b/docs/index.md
index 1a9cf01d1..f87a84d04 100644
--- a/docs/index.md
+++ b/docs/index.md
@@ -13,9 +13,9 @@ A collection of **local-first**, AI-powered command-line agents that run entirel
 Agent CLI provides a suite of powerful tools for voice and text interaction, designed for privacy, offline capability, and seamless integration with system-wide hotkeys and workflows.
 
 !!! important "Local and Private by Design"
-    All agents are designed to run **100% locally**. Your data—whether from your clipboard, microphone, or files—is never sent to any cloud API. This ensures your privacy and allows the tools to work completely offline.
+    All agents can run **100% locally**. Your data—whether from your clipboard, microphone, or files—stays on your machine unless you configure a cloud provider. This keeps workflows private and allows the tools to work offline.
 
-    You can also optionally configure the agents to use OpenAI/Gemini services.
+    You can optionally configure the agents to use OpenAI/Gemini services.
 
 ## Quick Demo
 
diff --git a/docs/installation/docker.md b/docs/installation/docker.md
index ff2daaef2..953e7f516 100644
--- a/docs/installation/docker.md
+++ b/docs/installation/docker.md
@@ -6,7 +6,8 @@ icon: lucide/container
 
 Universal Docker setup that works on any platform with Docker support.
 
-> **⚠️ Important Limitations**
+> [!WARNING]
+> **Important Limitations**
 >
 > - **macOS**: Docker does not support GPU acceleration. For 10x better performance, use [macOS native setup](macos.md)
 > - **Linux**: Limited GPU support. For full NVIDIA GPU acceleration, use [Linux native setup](linux.md)
@@ -37,7 +38,7 @@ Universal Docker setup that works on any platform with Docker support.
 3. **Install agent-cli:**
 
    ```bash
-   uv tools install agent-cli
+   uv tool install agent-cli
    # or: pip install agent-cli
    ```
 
diff --git a/docs/installation/index.md b/docs/installation/index.md
index 45eae050e..31198993a 100644
--- a/docs/installation/index.md
+++ b/docs/installation/index.md
@@ -15,7 +15,8 @@ Choose the best installation method for your platform and performance needs.
 | **NixOS**        | [System Integration](nixos.md) | ✅ NVIDIA GPU | Best        |
 | **Any Platform** | [Docker Setup](docker.md)      | ⚠️ Limited\*  | Good        |
 
-> **Note**: Docker on macOS does not support GPU acceleration. For best performance on Mac, use the native setup.
+> [!NOTE]
+> Docker on macOS does not support GPU acceleration. For best performance on Mac, use the [native setup](macos.md).
 
 ## Installation Methods
 
@@ -65,7 +66,7 @@ Choose the best installation method for your platform and performance needs.
 All installation methods set up these services:
 
 - **🧠 Ollama** - LLM server (gemma3:4b model)
-- **🎤 Wyoming Faster Whisper** - Speech-to-text
+- **🎤 Wyoming Whisper** - Speech-to-text (faster-whisper on Linux/Intel, MLX Whisper on Apple Silicon)
 - **🗣️ Wyoming Piper** - Text-to-speech
 - **👂 Wyoming OpenWakeWord** - Wake word detection
 
@@ -84,7 +85,7 @@ Once services are running, install the agent-cli package:
 
 ```bash
 # Using uv (recommended)
-uv tools install agent-cli
+uv tool install agent-cli
 
 # Using pip
 pip install agent-cli
diff --git a/docs/installation/linux.md b/docs/installation/linux.md
index c3793c4ba..51ac2ac64 100644
--- a/docs/installation/linux.md
+++ b/docs/installation/linux.md
@@ -6,8 +6,8 @@ icon: lucide/terminal
 
 Native Linux setup with full NVIDIA GPU acceleration for optimal performance.
 
-> **🐧 Recommended for Linux**
-> This setup provides optimal performance with full NVIDIA GPU acceleration support.
+> [!TIP]
+> **🐧 Recommended for Linux** — Optimal performance with full NVIDIA GPU acceleration.
 
 ## Prerequisites
 
@@ -41,7 +41,7 @@ Native Linux setup with full NVIDIA GPU acceleration for optimal performance.
 3. **Install agent-cli:**
 
    ```bash
-   uv tools install agent-cli
+   uv tool install agent-cli
    ```
 
 4. **Test the setup:**
diff --git a/docs/installation/macos.md b/docs/installation/macos.md
index e4a9174a4..5fcd6f87d 100644
--- a/docs/installation/macos.md
+++ b/docs/installation/macos.md
@@ -6,8 +6,8 @@ icon: lucide/apple
 
 Native macOS setup with full Metal GPU acceleration for optimal performance.
 
-> **🍎 Recommended for macOS**
-> This setup provides ~10x better performance than Docker by utilizing Metal GPU acceleration.
+> [!TIP]
+> **🍎 Recommended for macOS** — ~10x better performance than Docker via Metal GPU acceleration.
 
 ## Prerequisites
 
@@ -87,19 +87,25 @@ scripts/start-all-services.sh
 If you prefer running services individually:
 
 ```bash
-# Terminal 1: Ollama (native GPU acceleration)
+# Ollama (brew service recommended)
+brew services start ollama
+# Or run in foreground:
 ollama serve
 
-# Terminal 2: Whisper (CPU optimized)
+# Whisper (Apple Silicon: launchd service or manual)
+launchctl list com.wyoming_mlx_whisper
+# Or run in foreground:
 scripts/run-whisper.sh
 
-# Terminal 3: Piper (Apple Silicon compatible)
+# Piper
 scripts/run-piper.sh
 
-# Terminal 4: OpenWakeWord (macOS compatible fork)
+# OpenWakeWord
 scripts/run-openwakeword.sh
 ```
 
+Intel Macs: prefer Docker or a Linux-style Wyoming Faster Whisper setup; MLX Whisper is Apple Silicon only.
+
 ## Why Native Setup?
 
 - **10x faster than Docker** - Full Metal GPU acceleration
diff --git a/docs/installation/nixos.md b/docs/installation/nixos.md
index fe8f729b2..168f162b6 100644
--- a/docs/installation/nixos.md
+++ b/docs/installation/nixos.md
@@ -6,8 +6,8 @@ icon: lucide/snowflake
 
 Native NixOS setup using system configuration with full GPU acceleration support.
 
-> **❄️ For NixOS Users**
-> This setup integrates agent-cli services directly into your NixOS system configuration.
+> [!TIP]
+> **❄️ For NixOS Users** — Integrates agent-cli services directly into your NixOS system configuration.
 
 ## Prerequisites
 
diff --git a/docs/system-integration.md b/docs/system-integration.md
index 0a03164f6..c5a3de910 100644
--- a/docs/system-integration.md
+++ b/docs/system-integration.md
@@ -123,10 +123,10 @@ Install [AutoHotkey](https://www.autohotkey.com/) and create a script:
 
 ```ahk
 ; Transcribe to clipboard
-#+r::Run, wsl agent-cli transcribe --toggle --input-device-index 1
+#+r::Run, agent-cli transcribe --toggle --input-device-index 1
 
 ; Autocorrect clipboard
-#+a::Run, wsl agent-cli autocorrect
+#+a::Run, agent-cli autocorrect
 ```
 
 ### PowerToys