chore: update 9 file(s)

.gitignore

@@ -33,3 +33,5 @@ __pycache__/
 *.tmp
 temp/
 dist/
+.aider*
+*.iml

ARCHITECTURE.md

@@ -0,0 +1,72 @@
+User Input (Tab×2 or CLI)
+↓
+┌───────────────────────────────────────┐
+│ Entry Points                          │
+├───────────────────────────────────────┤
+│ • _finishsh() - Tab completion        │
+│ • command_command() - CLI mode        │
+└───────────────┬───────────────────────┘
+↓
+┌───────────────────────────────────────┐
+│ Configuration Layer                   │
+├───────────────────────────────────────┤
+│ acsh_load_config()                    │
+│ • Read ~/.finish/config               │
+│ • Set ACSH_PROVIDER, ACSH_ENDPOINT    │
+│ • Set ACSH_ACTIVE_API_KEY (if needed) │
+└───────────────┬───────────────────────┘
+↓
+┌───────────────────────────────────────┐
+│ Cache Layer                           │
+├───────────────────────────────────────┤
+│ • Check cache_dir/acsh-{hash}.txt     │
+│ • Return cached completions if exists │
+└───────────────┬───────────────────────┘
+↓ (cache miss)
+┌───────────────────────────────────────┐
+│ Context Builder                       │
+├───────────────────────────────────────┤
+│ _build_prompt()                       │
+│ • Command history (sanitized)         │
+│ • Terminal info (env vars, cwd)       │
+│ • Recent files (ls -ld)               │
+│ • Help message (cmd --help)           │
+└───────────────┬───────────────────────┘
+↓
+┌───────────────────────────────────────┐
+│ Payload Builder                       │
+├───────────────────────────────────────┤
+│ _build_payload()                      │
+│ • Format: {model, messages, temp}     │
+│ • Provider-specific options:          │
+│   - Ollama: {format:"json"}           │
+│   - LMStudio: {stream:true}           │
+│   - OpenAI: {response_format}         │
+└───────────────┬───────────────────────┘
+↓
+┌───────────────────────────────────────┐
+│ LLM Provider Layer                    │
+├───────────────────────────────────────┤
+│ openai_completion()                   │
+│ • curl to endpoint                    │
+│ • Ollama: no auth header              │
+│ • Others: Authorization header if key │
+│ • Parse response (JSON/streaming)     │
+└───────────────┬───────────────────────┘
+↓
+┌───────────────────────────────────────┐
+│ Response Parser                       │
+├───────────────────────────────────────┤
+│ • Extract completions array           │
+│ • Fallback parsing for non-JSON      │
+│ • Filter command-like lines           │
+└───────────────┬───────────────────────┘
+↓
+┌───────────────────────────────────────┐
+│ Output Layer                          │
+├───────────────────────────────────────┤
+│ • Write to cache                      │
+│ • Log usage (tokens, cost)            │
+│ • Return COMPREPLY array (completion) │
+│ • or stdout (CLI mode)                │
+└───────────────────────────────────────┘

Dockerfile

@@ -26,7 +26,7 @@ RUN apt-get update && \
 WORKDIR /opt/finish
 
 # Copy application files
-COPY finish.sh /usr/bin/finish
+COPY src/finish.sh /usr/bin/finish
 COPY README.md LICENSE ./
 
 # Make script executable

LICENSE

@@ -1,24 +0,0 @@
-BSD 2-Clause License
-
-Copyright (c) 2025, finish contributors
-
-Redistribution and use in source and binary forms, with or without
-modification, are permitted provided that the following conditions are met:
-
-1. Redistributions of source code must retain the above copyright notice, this
-   list of conditions and the following disclaimer.
-
-2. Redistributions in binary form must reproduce the above copyright notice,
-   this list of conditions and the following disclaimer in the documentation
-   and/or other materials provided with the distribution.
-
-THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
-AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
-IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
-DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE
-FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
-DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
-SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
-CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
-OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
-OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

README.md

@@ -1,300 +1,149 @@
-# finish.sh
+```markdown
+# finish
 
-Command-line completion powered by local LLMs. Press `Tab` twice and get intelligent suggestions based on your terminal context.
+AI-powered shell completion that runs 100% on your machine.
 
-## Quick Start
+One command and your terminal learns what you type next.
 
-### One-Line Install (Recommended)
+## Install
 
 ```bash
+# On Ubuntu/Debian, first ensure python3-venv is installed
+sudo apt-get install python3-venv
+
+# Then run the installer
 curl -sSL https://git.appmodel.nl/tour/finish/raw/branch/main/docs/install.sh | bash
+source ~/.bashrc # or ~/.zshrc
+
+# for plato special, add plato.lan to /etc/hosts:
+sudo nano /etc/hosts - 192.168.1.74 plato.lan
 ```
 
-### Manual Installation
+Press **Alt+\\** after typing any command to get intelligent completions—no cloud, no data leak, instant results.
+
+## How it works
+
+1. Captures your current directory, recent history, env vars, and available tools
+2. Analyzes your intent and builds a context-aware prompt
+3. Queries your local LLM (LM Studio, Ollama, or any OpenAI-compatible endpoint)
+4. Returns 2-5 ranked completions with an interactive picker
+5. Results are cached for instant replay
+
+## Usage
+
+Type a command, then press **Alt+\\**:
 
 ```bash
-git clone https://git.appmodel.nl/tour/finish.git finish
-cd finish
-./finish.sh install
-source ~/.bashrc
+# Natural language commands
+show gpu status                    # → nvidia-smi
+resolve title of website google.com # → curl -s https://google.com | grep -oP '<title>\K[^<]+'
+make file about dogs               # → echo "About Dogs" > dogs.md
+
+# Partial commands
+git commit                         # → git commit -m "..."
+docker run                         # → docker run -it --rm ubuntu bash
+find large files                   # → find . -type f -size +100M
 ```
 
-### Other Installation Methods
+Navigate with ↑↓, press Enter to accept, Esc to cancel.
 
-See [Installation](#installation) below for package managers, Docker, and more options.
+## Configure
 
-## What It Does
-
-finish.sh enhances your terminal with context-aware command suggestions. It analyzes:
-
-- Current directory and recent files
-- Command history
-- Environment variables
-- Command help output
-
-Then generates relevant completions using a local LLM.
-
-## Features
-
-**Local by Default**
-Runs with LM-Studio on localhost. No external API calls, no data leaving your machine.
-
-**Context-Aware**
-Understands what you're doing based on your environment and history.
-
-**Cached**
-Stores recent queries locally for instant responses.
-
-**Flexible**
-Switch between different models and providers easily.
-
-## Configuration
-
-View current settings:
+View current configuration:
 
 ```bash
 finish config
 ```
 
-Change the endpoint or model:
+Edit configuration file:
 
 ```bash
-finish config set endpoint http://plato.lan:1234/v1/chat/completions
-finish config set model your-model-name
+nano ~/.finish/finish.json
 ```
 
-Select a model interactively:
+## Configuration Examples
 
-```bash
-finish model
+### Local Ollama
+```json
+{
+  "provider": "ollama",
+  "model": "llama3:latest",
+  "endpoint": "http://localhost:11434/api/chat",
+  "temperature": 0.0,
+  "api_prompt_cost": 0.0,
+  "api_completion_cost": 0.0,
+  "max_history_commands": 20,
+  "max_recent_files": 20,
+  "cache_size": 100
+}
 ```
 
-## Usage
-
-Type a command and press `Tab` twice:
-
-```bash
-docker <TAB><TAB>
-git commit <TAB><TAB>
-ffmpeg <TAB><TAB>
+### LM Studio
+```json
+{
+  "provider": "lmstudio",
+  "model": "dolphin3.0-llama3.1-8b@q4_k_m",
+  "endpoint": "http://localhost:1234/v1/chat/completions",
+  "temperature": 0.0,
+  "api_prompt_cost": 0.0,
+  "api_completion_cost": 0.0,
+  "max_history_commands": 20,
+  "max_recent_files": 20,
+  "cache_size": 100
+}
 ```
 
-Natural language works too:
-
-```bash
-# find large files <TAB><TAB>
-# compress to zip <TAB><TAB>
-```
-
-Test it without executing:
-
-```bash
-finish command "your command"
-finish command --dry-run "your command"
-```
-
-## Installation
-
-### Quick Install (Linux/macOS)
-
-The fastest way to get started:
-
-```bash
-curl -sSL https://git.appmodel.nl/tour/finish/raw/branch/main/docs/install.sh | bash
-source ~/.bashrc  # or ~/.zshrc for zsh
-finish model
-```
-
-### Package Managers
-
-#### Homebrew (macOS)
-
-```bash
-brew tap appmodel/finish
-brew install finish
-finish install
-source ~/.bashrc
-```
-
-#### APT (Debian/Ubuntu)
-
-Download the `.deb` package from [releases](https://git.appmodel.nl/tour/finish/releases):
-
-```bash
-sudo dpkg -i finish_*.deb
-sudo apt-get install -f  # Install dependencies
-finish install
-source ~/.bashrc
-```
-
-### Docker
-
-Run finish.sh in a container:
-
-```bash
-# Build the image
-docker build -t finish .
-
-# Run interactively
-docker run -it finish
-
-# Or use docker-compose
-docker-compose up -d finish
-docker-compose exec finish bash
-```
-
-Inside the container:
-```bash
-finish install
-source ~/.bashrc
-finish model  # Configure your LLM endpoint
-```
-
-### From Source
-
-```bash
-git clone https://git.appmodel.nl/tour/finish.git finish
-cd finish
-chmod +x finish.sh
-sudo ln -s $PWD/finish.sh /usr/local/bin/finish
-finish install
-source ~/.bashrc
+### OpenAI (or compatible API)
+```json
+{
+  "provider": "lmstudio",
+  "model": "gpt-4",
+  "endpoint": "https://api.openai.com/v1/chat/completions",
+  "api_key": "sk-...",
+  "temperature": 0.0,
+  "api_prompt_cost": 0.03,
+  "api_completion_cost": 0.06,
+  "max_history_commands": 20,
+  "max_recent_files": 20,
+  "cache_size": 100
+}
 ```
 
 ## Requirements
 
-- Bash 4.0+ or Zsh 5.0+
-- curl
-- jq
-- bc
-- bash-completion (recommended)
-- LM-Studio or Ollama (for LLM inference)
+- Python 3.7+
+- Bash ≥4 or Zsh ≥5
+- A local LLM running (Ollama, LM Studio, etc.) or API access
 
-## Directory Structure
-
-```
-~/.finish/
-├── config              # Configuration file
-├── finish.log    # Usage log
-└── cache/              # Cached completions
-```
+Python dependencies (installed automatically):
+- httpx
+- prompt_toolkit
+- rich
 
 ## Commands
 
 ```bash
-finish install    # Set up finish
-finish remove     # Uninstall
-finish config     # Show/edit configuration
-finish model      # Select model
-finish enable     # Enable completions
-finish disable    # Disable completions
-finish clear      # Clear cache and logs
-finish usage      # Show usage statistics
-finish system     # Show system information
-finish --help     # Show help
+finish install    # Set up Alt+\ keybinding
+finish config     # Show current configuration
+finish command "text"  # Test completions manually
 ```
 
-## Providers
-
-Currently supports:
-
-- **LM-Studio** (default) - Local models via OpenAI-compatible API
-- **Ollama** - Local models via Ollama API
-
-Add custom providers by editing `finish.sh` and adding entries to `_finish_modellist`.
-
-## Development
-
-### Local Development
-
-Clone and link for development:
+## Advanced
 
+### Debug mode
 ```bash
-git clone https://git.appmodel.nl/tour/finish.git finish
-cd finish
-ln -s $PWD/finish.sh $HOME/.local/bin/finish
-finish install
+export FINISH_DEBUG=1
+finish command "your command here"
+cat ~/.finish/finish.log
 ```
 
-### Running Tests
-
+### Clear cache
 ```bash
-# Install BATS testing framework
-sudo apt-get install bats  # Ubuntu/Debian
-
-# Run tests
-./run_tests.sh
-
-# Or use Docker
-docker build -f Dockerfile.test -t finish:test .
-docker run --rm finish:test
+rm -rf ~/.finish/cache/*.json
 ```
 
-### Building Packages
-
-#### Debian Package
-
-```bash
-# Install build dependencies
-sudo apt-get install debhelper devscripts
-
-# Build the package
-dpkg-buildpackage -us -uc -b
-
-# Package will be created in parent directory
-ls ../*.deb
-```
-
-#### Docker Images
-
-```bash
-# Production image
-docker build -t finish:latest .
-
-# Test image
-docker build -f Dockerfile.test -t finish:test .
-
-# Using docker-compose
-docker-compose build
-```
-
-## Distribution & Publishing
-
-### Creating a Release
-
-1. Update version in `finish.sh` (ACSH_VERSION)
-2. Update version in `debian/changelog`
-3. Update version in `homebrew/finish.rb`
-4. Create and push a git tag:
-
-```bash
-git tag -a v0.5.0 -m "Release version 0.5.0"
-git push origin v0.5.0
-```
-
-This triggers the GitHub Actions workflow which:
-- Builds release artifacts
-- Creates Debian packages
-- Publishes Docker images to GitHub Container Registry
-- Creates a GitHub release with downloadable assets
-
-### Homebrew Formula
-
-After creating a release, update the Homebrew formula:
-
-1. Calculate SHA256 of the release tarball:
-   ```bash
-   curl -sL https://git.appmodel.nl/tour/finish/archive/v0.5.0.tar.gz | sha256sum
-   ```
-
-2. Update `homebrew/finish.rb` with the new SHA256 and version
-
-3. Submit to your Homebrew tap or the main Homebrew repository
-
-## Contributing
-
-Contributions are welcome! Please feel free to submit a Pull Request.
-
 ## License
 
-BSD 2-Clause License. See LICENSE file.
+BSD 2-Clause.
+
+```

debian/copyright

@@ -1,30 +0,0 @@
-Format: https://www.debian.org/doc/packaging-manuals/copyright-format/1.0/
-Upstream-Name: finish
-Upstream-Contact: https://git.appmodel.nl/tour/finish/finish
-Source: https://git.appmodel.nl/tour/finish
-
-Files: *
-Copyright: 2024-2025 finish.sh Contributors
-License: BSD-2-Clause
-
-License: BSD-2-Clause
- Redistribution and use in source and binary forms, with or without
- modification, are permitted provided that the following conditions are met:
- .
- 1. Redistributions of source code must retain the above copyright notice, this
-    list of conditions and the following disclaimer.
- .
- 2. Redistributions in binary form must reproduce the above copyright notice,
-    this list of conditions and the following disclaimer in the documentation
-    and/or other materials provided with the distribution.
- .
- THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
- AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
- IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
- DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE
- FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
- DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
- SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
- CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
- OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
- OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

docker-compose.yml

@@ -12,7 +12,7 @@ services:
     volumes:
       - finish-data:/root/.finish
     environment:
-      - ACSH_ENDPOINT=http://plato.lan:1234/v1/chat/completions
+      - ACSH_ENDPOINT=http://llm.plato/v1/chat/completions
     extra_hosts:
       - "host.docker.internal:host-gateway"
     command: /bin/bash

docs/install.sh

@@ -4,7 +4,7 @@
 
 set -e
 
-ACSH_VERSION="v0.5.1"
+ACSH_VERSION="v0.6.0"
 BRANCH_OR_VERSION=${1:-main}
 REPO_URL="https://git.appmodel.nl/tour/finish/raw/branch"
 
@@ -97,11 +97,9 @@ main() {
     SHELL_TYPE=$(detect_shell)
     case "$SHELL_TYPE" in
         zsh)
-            SCRIPT_NAME="finish.zsh"
             RC_FILE="$HOME/.zshrc"
             ;;
         bash)
-            SCRIPT_NAME="finish.sh"
             RC_FILE="$HOME/.bashrc"
             ;;
         *)
@@ -112,6 +110,29 @@ main() {
 
     echo "Detected shell: $SHELL_TYPE"
 
+    # Check Python and venv
+    echo "Checking Python installation..."
+    if ! command -v python3 > /dev/null 2>&1; then
+        echo_error "Python 3 is required but not found"
+        echo "Install it with:"
+        echo "  Ubuntu/Debian: sudo apt-get install python3 python3-pip python3-venv"
+        echo "  macOS: brew install python3"
+        exit 1
+    fi
+    echo_green "✓ Python 3 found: $(python3 --version)"
+
+    # Check if venv module is available
+    if ! python3 -m venv --help > /dev/null 2>&1; then
+        echo_error "python3-venv module is not installed"
+        echo "Install it with:"
+        echo "  Ubuntu/Debian: sudo apt-get install python3-venv"
+        echo "  CentOS/RHEL: sudo yum install python3-venv"
+        echo "  macOS: (included with python3)"
+        exit 1
+    fi
+    echo_green "✓ python3-venv available"
+    echo ""
+
     # Check dependencies
     echo "Checking dependencies..."
     if ! check_dependencies; then
@@ -144,17 +165,73 @@ main() {
         exit 1
     fi
 
-    # Create directory if needed
+    # Create directories if needed
     mkdir -p "$(dirname "$INSTALL_LOCATION")"
+    mkdir -p "$HOME/.venvs"
 
-    # Download script
-    echo "Downloading finish.sh..."
-    URL="$REPO_URL/$BRANCH_OR_VERSION/$SCRIPT_NAME"
+    # Download Python script
+    echo "Downloading finish.py..."
+    URL="$REPO_URL/$BRANCH_OR_VERSION/src/finish.py"
     if ! download_file "$URL" "$INSTALL_LOCATION"; then
         echo_error "Failed to download from $URL"
         exit 1
     fi
 
+    # Download requirements.txt
+    echo "Downloading requirements.txt..."
+    TEMP_REQ="/tmp/finish_requirements.txt"
+    REQ_URL="$REPO_URL/$BRANCH_OR_VERSION/requirements.txt"
+    if ! download_file "$REQ_URL" "$TEMP_REQ"; then
+        echo_error "Failed to download requirements.txt from $REQ_URL"
+        exit 1
+    fi
+
+    # Create virtualenv and install dependencies
+    echo "Creating virtual environment..."
+    VENV_PATH="$HOME/.venvs/finish"
+    if [ -d "$VENV_PATH" ]; then
+        echo "Virtual environment already exists, removing old one..."
+        rm -rf "$VENV_PATH"
+    fi
+
+    if ! python3 -m venv "$VENV_PATH"; then
+        echo_error "Failed to create virtual environment"
+        echo "Make sure python3-venv is installed:"
+        echo "  Ubuntu/Debian: sudo apt-get install python3-venv"
+        exit 1
+    fi
+    echo_green "✓ Virtual environment created at $VENV_PATH"
+
+    echo "Installing Python dependencies..."
+    if ! "$VENV_PATH/bin/pip" install --quiet --upgrade pip; then
+        echo_error "Failed to upgrade pip"
+        exit 1
+    fi
+
+    if ! "$VENV_PATH/bin/pip" install --quiet -r "$TEMP_REQ"; then
+        echo_error "Failed to install dependencies"
+        cat "$TEMP_REQ"
+        exit 1
+    fi
+    echo_green "✓ Dependencies installed"
+    rm -f "$TEMP_REQ"
+
+    # Update shebang to use venv python (compatible with macOS and Linux)
+    if sed --version >/dev/null 2>&1; then
+        # GNU sed (Linux)
+        sed -i "1s|.*|#!$VENV_PATH/bin/python3|" "$INSTALL_LOCATION"
+    else
+        # BSD sed (macOS)
+        sed -i '' "1s|.*|#!$VENV_PATH/bin/python3|" "$INSTALL_LOCATION"
+    fi
+
+    # Verify shebang was updated correctly
+    SHEBANG=$(head -n 1 "$INSTALL_LOCATION")
+    if [ "$SHEBANG" != "#!$VENV_PATH/bin/python3" ]; then
+        echo_error "Failed to update shebang. Got: $SHEBANG"
+        exit 1
+    fi
+
     chmod +x "$INSTALL_LOCATION"
     echo_green "✓ Installed to $INSTALL_LOCATION"
     echo ""
@@ -183,8 +260,18 @@ main() {
         fi
     fi
 
-    # Run finish install
-    echo "Running finish installation..."
+    # Test that finish works
+    echo "Testing installation..."
+    if ! "$INSTALL_LOCATION" --version > /dev/null 2>&1; then
+        echo_error "finish command failed to execute"
+        echo "Testing with direct python call..."
+        "$VENV_PATH/bin/python3" "$INSTALL_LOCATION" --version
+        exit 1
+    fi
+    echo_green "✓ finish command works"
+
+    # Run finish install to set up keybinding
+    echo "Setting up shell keybinding..."
     if "$INSTALL_LOCATION" install; then
         echo ""
         echo_green "=========================================="
@@ -195,10 +282,18 @@ main() {
         echo "  1. Reload your shell configuration:"
         echo "     source $RC_FILE"
         echo ""
-        echo "  2. Select a language model:"
-        echo "     finish model"
+        echo "  2. Configure your LLM endpoint:"
+        echo "     finish config"
         echo ""
-        echo "  3. Start using by pressing Tab twice after any command"
+        echo "  3. Edit config if needed:"
+        echo "     nano ~/.finish/finish.json"
+        echo ""
+        echo "  4. Start using by pressing Alt+\\ after typing a command"
+        echo ""
+        echo "Example:"
+        echo "  Type: show gpu status"
+        echo "  Press: Alt+\\"
+        echo "  Select completion with ↑↓ arrows, Enter to accept"
         echo ""
         echo "Documentation: https://git.appmodel.nl/tour/finish"
     else

docs/uninstall.sh

@@ -0,0 +1,42 @@
+#!/bin/sh
+# finish uninstaller
+
+echo "Removing finish installation..."
+
+# Remove binary
+if [ -f "$HOME/.local/bin/finish" ]; then
+    rm -f "$HOME/.local/bin/finish"
+    echo "✓ Removed $HOME/.local/bin/finish"
+fi
+
+# Remove venv
+if [ -d "$HOME/.venvs/finish" ]; then
+    rm -rf "$HOME/.venvs/finish"
+    echo "✓ Removed $HOME/.venvs/finish"
+fi
+
+# Remove config and cache
+if [ -d "$HOME/.finish" ]; then
+    rm -rf "$HOME/.finish"
+    echo "✓ Removed $HOME/.finish"
+fi
+
+# Remove keybinding from .bashrc
+if [ -f "$HOME/.bashrc" ]; then
+    if grep -q "# finish.py key-binding" "$HOME/.bashrc"; then
+        sed -i '/# finish.py key-binding/,+8d' "$HOME/.bashrc"
+        echo "✓ Removed keybinding from ~/.bashrc"
+    fi
+fi
+
+# Remove keybinding from .zshrc
+if [ -f "$HOME/.zshrc" ]; then
+    if grep -q "# finish.py key-binding" "$HOME/.zshrc"; then
+        sed -i '/# finish.py key-binding/,+8d' "$HOME/.zshrc"
+        echo "✓ Removed keybinding from ~/.zshrc"
+    fi
+fi
+
+echo ""
+echo "Uninstall complete!"
+echo "Run 'source ~/.bashrc' to reload your shell"

finish.iml

@@ -1,12 +0,0 @@
-<?xml version="1.0" encoding="UTF-8"?>
-<module type="WEB_MODULE" version="4">
-  <component name="NewModuleRootManager" inherit-compiler-output="true">
-    <exclude-output />
-    <content url="file://$MODULE_DIR$">
-      <excludeFolder url="file://$MODULE_DIR$/.git" />
-      <excludeFolder url="file://$MODULE_DIR$/tests" />
-    </content>
-    <orderEntry type="inheritedJdk" />
-    <orderEntry type="sourceFolder" forTests="false" />
-  </component>
-</module>

requirements.txt

@@ -0,0 +1,3 @@
+httpx
+prompt_toolkit
+rich

src/finish.py

@@ -0,0 +1,442 @@
+#!/usr/bin/env python3
+"""
+finish.py - AI shell completions that never leave your machine.
+"""
+from __future__ import annotations
+import hashlib
+
+import argparse, asyncio, json, os, re, shlex, subprocess, sys, time
+from pathlib import Path
+from typing import List, Dict, Optional
+
+import httpx
+from prompt_toolkit import ANSI, Application
+from prompt_toolkit.buffer import Buffer
+from prompt_toolkit.key_binding import KeyBindings
+from prompt_toolkit.layout import (
+    ConditionalContainer, FormattedTextControl, HSplit, Layout, Window
+)
+from prompt_toolkit.widgets import Label
+from rich.console import Console
+from rich.spinner import Spinner
+
+VERSION = "0.5.0"
+CFG_DIR   = Path.home()/".finish"
+CFG_FILE  = CFG_DIR/"finish.json"
+CACHE_DIR = CFG_DIR/"cache"
+LOG_FILE  = CFG_DIR/"finish.log"
+
+console = Console()
+
+# --------------------------------------------------------------------------- #
+#  Config
+# --------------------------------------------------------------------------- #
+DEFAULT_CFG = {
+    "provider": "ollama",
+    "model": "llama3:latest",
+    "endpoint": os.getenv("LLM_FINISH_URL", "http://localhost:11434/api/chat"),
+    "temperature": 0.0,
+    "api_prompt_cost": 0.0,
+    "api_completion_cost": 0.0,
+    "max_history_commands": 20,
+    "max_recent_files": 20,
+    "cache_size": 100,
+}
+
+def cfg() -> dict:
+    if not CFG_FILE.exists():
+        CFG_DIR.mkdir(exist_ok=True)
+        CFG_FILE.write_text(json.dumps(DEFAULT_CFG, indent=2))
+    return json.loads(CFG_FILE.read_text())
+
+# --------------------------------------------------------------------------- #
+#  Context builders
+# --------------------------------------------------------------------------- #
+def _sanitise_history() -> str:
+    hist = subprocess.check_output(["bash", "-ic", "history"]).decode()
+    # scrub tokens / hashes
+    for pat in [
+        r"\b[0-9a-f]{32,40}\b",          # long hex
+        r"\b[A-Za-z0-9-]{36}\b",         # uuid
+        r"\b[A-Za-z0-9]{16,40}\b",       # api-keyish
+    ]:
+        hist = re.sub(pat, "REDACTED", hist)
+    return "\n".join(hist.splitlines()[-cfg()["max_history_commands"]:])
+
+def _recent_files() -> str:
+    try:
+        files = subprocess.check_output(
+            ["find", ".", "-maxdepth", "1", "-type", "f", "-printf", "%T@ %p\n"],
+            stderr=subprocess.DEVNULL,
+        ).decode()
+        return "\n".join(sorted(files.splitlines(), reverse=True)[: cfg()["max_recent_files"]])
+    except Exception:
+        return ""
+
+def _installed_tools() -> str:
+    """Check for commonly used tools that might be relevant"""
+    tools = ["curl", "wget", "jq", "grep", "awk", "sed", "python3", "node", "docker", "git", "nvcc", "nvidia-smi"]
+    available = []
+    for tool in tools:
+        try:
+            subprocess.run(["which", tool], capture_output=True, timeout=0.5, check=True)
+            available.append(tool)
+        except Exception:
+            pass
+    return ", ".join(available) if available else "standard shell tools"
+
+def _get_context_info() -> dict:
+    """Gather comprehensive shell context"""
+    return {
+        "user": os.getenv("USER", "unknown"),
+        "pwd": os.getcwd(),
+        "home": os.getenv("HOME", ""),
+        "hostname": os.getenv("HOSTNAME", os.getenv("HOST", "localhost")),
+        "shell": os.getenv("SHELL", "bash"),
+        "tools": _installed_tools(),
+    }
+
+def build_prompt(user_input: str) -> str:
+    """Build an enhanced prompt with better context and instructions"""
+    c = cfg()
+    ctx = _get_context_info()
+
+    # Analyze user intent
+    user_words = user_input.lower().split()
+    intent_hints = []
+    if any(word in user_words for word in ["resolve", "get", "fetch", "download", "curl", "wget"]):
+        intent_hints.append("The user wants to fetch/download data")
+    if any(word in user_words for word in ["website", "url", "http", "html", "title"]):
+        intent_hints.append("The user is working with web content")
+    if any(word in user_words for word in ["parse", "extract", "grep", "find"]):
+        intent_hints.append("The user wants to extract/parse data")
+    if any(word in user_words for word in ["create", "make", "new", "write", "edit"]):
+        intent_hints.append("The user wants to create or edit files")
+    if any(word in user_words for word in ["file", "document", "text", "story", "about"]):
+        intent_hints.append("The user is working with files/documents")
+    if any(word in user_words for word in ["install", "setup", "configure", "apt", "pip"]):
+        intent_hints.append("The user wants to install or configure software")
+    if any(word in user_words for word in ["gpu", "nvidia", "cuda", "memory", "cpu"]):
+        intent_hints.append("The user wants system/hardware information")
+
+    intent_context = "\n".join(f"- {hint}" for hint in intent_hints) if intent_hints else "- General command completion"
+
+    prompt = f"""You are an intelligent bash completion assistant. Analyze the user's intent and provide practical, executable commands.
+
+USER INPUT: {user_input}
+
+DETECTED INTENT:
+{intent_context}
+
+SHELL CONTEXT:
+- User: {ctx['user']}
+- Working directory: {ctx['pwd']}
+- Hostname: {ctx['hostname']}
+- Available tools: {ctx['tools']}
+
+RECENT COMMAND HISTORY:
+{_sanitise_history()}
+
+INSTRUCTIONS:
+1. Understand what the user wants to accomplish (not just the literal words)
+2. Generate 2-5 practical bash commands that achieve the user's goal
+3. Prefer common tools (curl, wget, grep, awk, sed, jq, python)
+4. For file creation: use echo with heredoc, cat, or nano/vim
+5. For web tasks: use curl with headers, pipe to grep/sed/awk for parsing
+6. For complex parsing: suggest one-liners with proper escaping
+7. Commands should be copy-paste ready and executable
+8. Order by most likely to least likely intent
+
+EXAMPLES:
+- "resolve title of website example.com" → curl -s https://example.com | grep -oP '<title>\\K[^<]+'
+- "show gpu status" → nvidia-smi
+- "download json from api.com" → curl -s https://api.com/data | jq .
+- "make a new file story.txt" → nano story.txt
+- "create file and write hello" → echo "hello" > file.txt
+- "write about dogs to file" → cat > dogs.txt << 'EOF' (then Ctrl+D to finish)
+
+OUTPUT FORMAT (JSON only, no other text):
+{{"completions":["command1", "command2", "command3"]}}"""
+
+    return prompt
+
+# --------------------------------------------------------------------------- #
+#  LLM call
+# --------------------------------------------------------------------------- #
+async def llm_complete(prompt: str) -> List[str]:
+    c = cfg()
+    payload = {
+        "model": c["model"],
+        "temperature": c["temperature"],
+        "messages": [
+            {"role": "system", "content": "You are a helpful bash completion assistant."},
+            {"role": "user", "content": prompt},
+        ],
+        "response_format": {"type": "text"},
+    }
+    if c["provider"] == "ollama":
+        payload["format"] = "json"
+        payload["stream"] = False
+
+    headers = {"Content-Type": "application/json"}
+    if c.get("api_key"):
+        headers["Authorization"] = f"Bearer {c['api_key']}"
+
+    try:
+        async with httpx.AsyncClient(timeout=30) as client:
+            resp = await client.post(c["endpoint"], headers=headers, json=payload)
+            resp.raise_for_status()
+            body = resp.json()
+    except httpx.ConnectError as e:
+        console.print(f"[red]Error:[/] Cannot connect to {c['endpoint']}")
+        console.print(f"[yellow]Check that your LLM server is running and endpoint is correct.[/]")
+        console.print(f"[yellow]Run 'finish config' to see current config.[/]")
+        return []
+    except Exception as e:
+        console.print(f"[red]Error:[/] {e}")
+        return []
+
+    # extract content
+    if c["provider"] == "ollama":
+        raw = body["message"]["content"]
+    else:
+        raw = body["choices"][0]["message"]["content"]
+
+    # Debug logging
+    if os.getenv("FINISH_DEBUG"):
+        LOG_FILE.parent.mkdir(exist_ok=True)
+        with open(LOG_FILE, "a") as f:
+            f.write(f"\n=== {time.strftime('%Y-%m-%d %H:%M:%S')} ===\n")
+            f.write(f"Raw response:\n{raw}\n")
+
+    # try json first - use strict=False to be more lenient
+    try:
+        result = json.loads(raw, strict=False)["completions"]
+        if os.getenv("FINISH_DEBUG"):
+            with open(LOG_FILE, "a") as f:
+                f.write(f"Parsed completions: {result}\n")
+        return result
+    except Exception as e:
+        if os.getenv("FINISH_DEBUG"):
+            with open(LOG_FILE, "a") as f:
+                f.write(f"JSON parse failed: {e}\n")
+                f.write(f"Attempting manual extraction...\n")
+
+        # Try to extract JSON array manually using regex
+        try:
+            match = re.search(r'"completions"\s*:\s*\[(.*?)\]', raw, re.DOTALL)
+            if match:
+                # Extract commands from the array - handle escaped quotes
+                array_content = match.group(1)
+                # Find all quoted strings
+                commands = re.findall(r'"([^"\\]*(?:\\.[^"\\]*)*)"', array_content)
+                if commands:
+                    if os.getenv("FINISH_DEBUG"):
+                        with open(LOG_FILE, "a") as f:
+                            f.write(f"Manual extraction succeeded: {commands}\n")
+                    return commands[:5]
+        except Exception as e2:
+            if os.getenv("FINISH_DEBUG"):
+                with open(LOG_FILE, "a") as f:
+                    f.write(f"Manual extraction failed: {e2}\n")
+
+        # fallback: grep command-like lines
+        fallback = [ln for ln in raw.splitlines() if re.match(r"^(ls|cd|find|cat|grep|echo|mkdir|rm|cp|mv|pwd|chmod|chown|nano|vim|touch)\b", ln)][:5]
+        if os.getenv("FINISH_DEBUG"):
+            with open(LOG_FILE, "a") as f:
+                f.write(f"Fallback completions: {fallback}\n")
+        return fallback
+
+# --------------------------------------------------------------------------- #
+#  TUI picker
+# --------------------------------------------------------------------------- #
+async def select_completion(completions: List[str]) -> Optional[str]:
+    if not completions:
+        return None
+    if len(completions) == 1:
+        return completions[0]
+
+    # If not in an interactive terminal, return first completion
+    if not sys.stdin.isatty():
+        return completions[0]
+
+    # Import here to avoid issues when not needed
+    from prompt_toolkit.input import create_input
+    from prompt_toolkit.output import create_output
+
+    kb = KeyBindings()
+    current = 0
+
+    def get_text():
+        return [
+            ("", "Select completion  (↑↓ navigate, Enter accept, Esc cancel)\n\n"),
+            *[
+                ("[SetCursorPosition]" if i == current else "", f"{c}\n")
+                for i, c in enumerate(completions)
+            ],
+        ]
+
+    @kb.add("up")
+    def _up(event):
+        nonlocal current
+        current = (current - 1) % len(completions)
+
+    @kb.add("down")
+    def _down(event):
+        nonlocal current
+        current = (current + 1) % len(completions)
+
+    @kb.add("enter")
+    def _accept(event):
+        event.app.exit(result=completions[current])
+
+    @kb.add("escape")
+    def _cancel(event):
+        event.app.exit(result=None)
+
+    control = FormattedTextControl(get_text)
+
+    # Force output to /dev/tty to avoid interfering with bash command substitution
+    try:
+        output = create_output(stdout=open('/dev/tty', 'w'))
+        input_obj = create_input(stdin=open('/dev/tty', 'r'))
+    except Exception:
+        # Fallback to first completion if tty not available
+        return completions[0]
+
+    app = Application(
+        layout=Layout(HSplit([Window(control, height=len(completions) + 2)])),
+        key_bindings=kb,
+        mouse_support=False,
+        erase_when_done=True,
+        output=output,
+        input=input_obj,
+    )
+    return await app.run_async()
+
+# --------------------------------------------------------------------------- #
+#  Cache
+# --------------------------------------------------------------------------- #
+def cached(key: str) -> Optional[List[str]]:
+    if not CACHE_DIR.exists():
+        CACHE_DIR.mkdir(parents=True)
+    f = CACHE_DIR / f"{key}.json"
+    if f.exists():
+        return json.loads(f.read_text())
+    return None
+
+def store_cache(key: str, completions: List[str]) -> None:
+    f = CACHE_DIR / f"{key}.json"
+    f.write_text(json.dumps(completions))
+    # LRU eviction
+    all_files = sorted(CACHE_DIR.glob("*.json"), key=lambda p: p.stat().st_mtime)
+    for old in all_files[: -cfg()["cache_size"]]:
+        old.unlink(missing_ok=True)
+
+# --------------------------------------------------------------------------- #
+#  Main entry
+# --------------------------------------------------------------------------- #
+async def complete_line(line: str) -> Optional[str]:
+    key = hashlib.md5(line.encode()).hexdigest()
+    comps = cached(key)
+    if comps is None:
+        # Write to /dev/tty if available to avoid interfering with command substitution
+        try:
+            status_console = Console(file=open('/dev/tty', 'w'), stderr=False)
+        except Exception:
+            status_console = console
+
+        with status_console.status("[green]Thinking…[/]"):
+            comps = await llm_complete(build_prompt(line))
+        store_cache(key, comps)
+    return await select_completion(comps)
+
+# --------------------------------------------------------------------------- #
+#  CLI
+# --------------------------------------------------------------------------- #
+def install_keybinding():
+    r"""Inject Bash binding Alt+\ -> finish"""
+    rc = Path.home()/".bashrc"
+    marker = "# finish.py key-binding"
+
+    # Use bind -x to allow modifying READLINE_LINE
+    snippet = f'''{marker}
+_finish_complete() {{
+    local result
+    result=$(finish --readline-complete "$READLINE_LINE" 2>/dev/null)
+    if [[ -n "$result" ]]; then
+        READLINE_LINE="$result"
+        READLINE_POINT=${{#READLINE_LINE}}
+    fi
+}}
+bind -x '"\\e\\\\": _finish_complete'
+'''
+
+    text = rc.read_text() if rc.exists() else ""
+    if marker in text:
+        return
+    rc.write_text(text + "\n" + snippet)
+    console.print("[green]Key-binding installed (Alt+\\)[/] - restart your shell.")
+
+def main():
+    # Handle readline-complete flag before argparse (for bash bind -x)
+    if "--readline-complete" in sys.argv:
+        idx = sys.argv.index("--readline-complete")
+        line = sys.argv[idx + 1] if idx + 1 < len(sys.argv) else ""
+        if line.strip():
+            choice = asyncio.run(complete_line(line))
+            if choice:
+                print(choice)
+        return
+
+    # Legacy flag support
+    if sys.argv[-1] == "--accept-current-line":
+        line = os.environ.get("READLINE_LINE", "")
+        if line.strip():
+            choice = asyncio.run(complete_line(line))
+            if choice:
+                sys.stdout.write(f"\x1b]0;\a")
+                sys.stdout.write(f"\x1b[2K\r")
+                sys.stdout.write(choice)
+                sys.stdout.flush()
+        return
+
+    parser = argparse.ArgumentParser(prog="finish", description="AI shell completions")
+    parser.add_argument("--version", action="version", version=VERSION)
+    sub = parser.add_subparsers(dest="cmd")
+    sub.add_parser("install", help="add Alt+\\ key-binding to ~/.bashrc")
+    sub.add_parser("config", help="show current config")
+    p = sub.add_parser("command", help="simulate double-tab")
+    p.add_argument("words", nargs="*", help="partial command")
+    p.add_argument("--dry-run", action="store_true", help="show prompt only")
+    args = parser.parse_args()
+
+    if args.cmd == "install":
+        install_keybinding()
+        return
+    if args.cmd == "config":
+        console.print_json(json.dumps(cfg()))
+        return
+    if args.cmd == "command":
+        line = " ".join(args.words)
+        if args.dry_run:
+            console.print(build_prompt(line))
+            return
+        choice = asyncio.run(complete_line(line))
+        if choice:
+            # Check if we're in an interactive shell
+            if os.isatty(sys.stdout.fileno()):
+                console.print(f"[dim]Selected:[/] [cyan]{choice}[/]")
+                console.print(f"\n[yellow]Tip:[/] Use [bold]Alt+\\[/] keybinding for seamless completion!")
+            else:
+                print(choice)
+        return
+    if len(sys.argv) == 1:
+        parser.print_help()
+        return
+
+if __name__ == "__main__":
+    try:
+        main()
+    except KeyboardInterrupt:
+        pass