Windows-first semantic command shell with AI reasoning, self-healing execution, and vector memory.
Natural language interface for Windows with LLM-powered intent routing, automatic error recovery, and semantic command history.
- Natural Language Commands - "open brave", "list 20 processes", "what did I do yesterday?"
- AI-Powered Routing - Ollama integration for understanding conversational queries
- Self-Healing Execution - Automatic error detection and repair with human-in-the-loop
- Semantic Memory - Vector-based command history search (optional ChromaDB)
- Clipboard History - Persistent clipboard manager with search and restore
- Smart Autocomplete - Context-aware tab completion with frequency tracking and previews
- Tab Completion - Intelligent autocomplete that learns from your usage patterns
- Process Management - List, monitor, and manage Windows processes
- File Operations - Navigate folders, list files, watch for changes
- Application Launcher - Open any Windows application by name
# Install
pip install -e .
# Optional: Full features (semantic memory, notifications, etc.)
pip install -e ".[full]"
# Run
ishellintellishell> open desktop
intellishell> list 5 most recent items in downloads
intellishell> watch downloads for pdfintellishell> list 20 processes
intellishell> kill notepad
intellishell> check adminintellishell> open brave
intellishell> open discord
intellishell> open cursor
intellishell> open notepadintellishell> what are my recent commands
intellishell> show me 15 processes currently running
intellishell> list 7 files in my downloads folderintellishell> system info
intellishell> get hostname
intellishell> disk spaceUser Input
|
v
[Natural Language Detection]
|
+---> [LLM Router (Ollama)] ---> Intent Match
|
+---> [Rule-Based Parser] ------> Intent Match
|
v
[Execution Planner]
|
v
[Self-Healing Executor]
|
+---> Try Execute
|
+---> Detect Error
|
+---> AI Repair Suggestion
|
+---> Human Approval
|
+---> Retry
|
v
[Provider Execution]
|
+---> FileSystem Provider
+---> App Provider
+---> System Provider
+---> Watch Provider
+---> Doctor Provider
+---> Memory Provider
+---> Clipboard Provider
IntelliShell uses Ollama for natural language understanding:
# Natural language queries are automatically routed to LLM
intellishell> show me my computer name
# LLM interprets -> get_hostname intent
intellishell> what processes are using the most memory
# LLM interprets -> list_processes intentSetup:
- Install Ollama: https://ollama.ai
- Pull a model:
ollama pull llama3:8b - IntelliShell auto-detects and uses it
Disable AI: ishell --no-ai
Automatic error recovery with AI-powered repair suggestions:
intellishell> opn desktp
# Detects typo, suggests correction
# "Did you mean: open desktop?"When commands fail, the self-healing executor:
- Detects the error type
- Generates repair suggestion via LLM
- Asks for human approval
- Retries with corrected command
Disable self-healing: ishell --no-self-healing
Vector-based command history for semantic search:
# Install ChromaDB
pip install chromadb
# Then use semantic queries
intellishell> what folder did I open yesterday?
intellishell> what did I do with downloads?
intellishell> recent memoriesCommands are automatically indexed as vector embeddings in ~/.intellishell/vector_store.
Disable memory: ishell --no-memory
Persistent clipboard history with automatic monitoring and search:
# View clipboard history (last 20 entries)
intellishell> clipboard history
# Search clipboard history
intellishell> clipboard search "that API key"
intellishell> clipboard search password
# Restore previous clipboard entry
intellishell> clipboard restore 5
# View statistics
intellishell> clipboard stats
# Control monitoring
intellishell> clipboard start monitoring
intellishell> clipboard stop monitoring
# Clear history
intellishell> clipboard clearFeatures:
- Automatic background monitoring of clipboard changes
- Persistent storage to
~/.intellishell/clipboard_history.jsonl - Search through clipboard history
- Restore any previous clipboard entry
- Deduplication (skips consecutive duplicates)
- Size limits (default: 100 entries, 10MB max)
- Thread-safe operations
Storage Location: ~/.intellishell/clipboard_history.jsonl
Intelligent tab completion that learns from your usage patterns and provides contextual suggestions:
Features:
- Frequency-Based Ranking - Most-used commands appear first
- Recency Scoring - Recently used commands get priority
- Time-of-Day Awareness - Suggests commands you typically use at this hour
- Command Sequence Learning - Predicts next command based on previous one
- Fuzzy Matching - Handles typos and partial matches
- Preview Generation - Shows what each command does
- Usage Statistics - Displays how often you've used each command
How It Works:
# Press TAB for completions
intellishell> op[TAB]
# Shows: open desktop (used 15x) | Usually used around 9am
# open downloads (used 8x)
# open documents (used 3x)
# Fuzzy matching handles typos
intellishell> lst[TAB]
# Shows: list downloads
# list desktop
# list files
# Context-aware suggestions
intellishell> open desktop
intellishell> [TAB]
# Shows: list files (often follows "open desktop")
# list desktopLearning:
- Automatically tracks command frequency
- Records time-of-day patterns
- Learns command sequences
- Persists to
~/.intellishell/completion_stats.json
Privacy:
- All data stored locally
- No telemetry or cloud sync
- Can be cleared anytime
open desktop- Open Desktop folderopen downloads- Open Downloads folderopen documents- Open Documents folderopen recycle bin- Open Recycle Binopen explorer- Open File Explorerlist files- List files in current directorylist downloads- List files in Downloadslist desktop- List files on Desktop
open notepad- Launch Notepadopen calculator- Launch Calculatoropen settings- Launch Windows Settingsopen task manager- Launch Task Manageropen control panel- Launch Control Panelopen [app name]- Launch any application (brave, discord, chrome, cursor, etc.)
list processes- List top 10 processes by memorylist [N] processes- List N processeskill process [PID]- Kill process by PIDkill [name]- Kill process by namemost memory- Show top memory consumercheck admin- Check if running as Administrator
system info- Display system informationget hostname- Show computer nameget username- Show current userdisk space- Show disk usage
watch downloads- Monitor Downloads folderwatch downloads for pdf- Monitor for specific file typelist watches- Show active file monitorsstop watching- Stop all file monitors
check system health- Run system diagnosticscheck dependencies- Verify installed dependencies
what did I [action]- Semantic search of command historyrecent memories- Show recent command memories
clipboard history- Show clipboard history (last 20 entries)clipboard search <query>- Search clipboard historyclipboard restore N- Restore entry N to clipboardclipboard clear- Clear clipboard historyclipboard stats- Show clipboard statisticsclipboard start monitoring- Start background clipboard monitoringclipboard stop monitoring- Stop clipboard monitoring
help- Show available commandshistory- Show command history!N- Replay command N from historystats- Show session statisticsmanifest- Show detailed system manifestclear/cls- Clear screenexit/quit- Exit shell
# Interactive mode (default)
ishell
# Single command execution
ishell -c "open desktop"
# Debug mode (show intent matching details)
ishell --debug
# Disable AI bridge
ishell --no-ai
# Disable semantic memory
ishell --no-memory
# Custom log file
ishell --log-file /path/to/log.txt
# Show version
ishell --version~/.intellishell/
├── vector_store/ # ChromaDB vector storage (if enabled)
├── logs/
│ └── shell.log # Structured logs
├── clipboard_history.jsonl # Clipboard history storage
├── completion_stats.json # Smart autocomplete learning data
├── history.jsonl # Transaction log
└── repairs.jsonl # Self-healing repair log
Every command is logged to ~/.intellishell/history.jsonl:
{
"timestamp": "2026-01-16T10:30:45.123456",
"user_input": "open desktop",
"intent_name": "open_desktop",
"provider_name": "filesystem",
"confidence": 0.95,
"success": true,
"entities": [],
"metadata": {}
}pydantic>=2.0.0- Data validationrequests>=2.28.0- HTTP clientprompt-toolkit>=3.0.0- Interactive shell with tab completion
psutil>=5.9.0- Process management, disk spacepyperclip>=1.8.0- Clipboard integrationrich>=13.0.0- Enhanced terminal UIwatchdog>=3.0.0- Filesystem monitoringplyer>=2.1.0orwin10toast>=0.9- Native notificationschromadb>=0.4.0- Semantic memory (vector storage)
Install all features:
pip install -e ".[full]"- Read-Only Default - Most operations are non-destructive
- Critical Process Protection - Cannot kill system processes
- Admin Detection - Warns when elevated privileges needed
- Human-in-the-Loop - Destructive actions require confirmation
- Safety Levels - GREEN (safe), YELLOW (caution), RED (requires approval)
- Circuit Breaker - Prevents repeated failures
- Parser: <50ms intent matching
- Entity Extraction: <10ms
- LLM Routing: ~100-500ms (when needed)
- Command Execution: Async, non-blocking
from intellishell.providers.base import BaseProvider, IntentTrigger, ExecutionResult
class MyProvider(BaseProvider):
@property
def name(self) -> str:
return "my_provider"
@property
def description(self) -> str:
return "My custom provider"
def _initialize_triggers(self) -> None:
self.triggers = [
IntentTrigger(
pattern="my command",
intent_name="do_thing",
weight=1.0,
aliases=["alternative command"]
)
]
async def execute(self, intent_name, context=None):
if intent_name == "do_thing":
return ExecutionResult(
success=True,
message="Thing done!",
data={"result": "success"}
)Register in intellishell/providers/registry.py:
from intellishell.providers.my_provider import MyProvider
def auto_discover(self, semantic_memory=None):
providers = [
# ... existing providers ...
MyProvider(),
]# Run all tests
pytest tests/ -v
# Specific test files
pytest tests/test_parser.py -v
pytest tests/test_providers.py -v
pytest tests/test_session.py -vpip install chromadb- Install Ollama: https://ollama.ai
- Pull a model:
ollama pull llama3:8b - Verify:
ollama list
- Requires
prompt-toolkit>=3.0.0 - Install:
pip install prompt-toolkit
Some commands (like killing processes) require Administrator privileges:
- Right-click terminal
- Select "Run as Administrator"
- Run
ishellagain
MIT
- Fork the repository
- Create a feature branch
- Implement your provider/feature
- Add tests
- Submit a pull request
- Plugin system for external providers
- Cross-platform support (Linux, macOS)
- Voice command integration
- Multi-language support
- Remote providers via RPC
- Advanced process monitoring and analytics
- Integration with Windows Terminal
- Custom theme support