Package Exports

This package does not declare an exports field, so the exports above have been automatically detected and optimized by JSPM instead. If any package subpath is missing, it is recommended to post an issue to the original package (@aiiware/aii) to support the "exports" field. If that is not possible, create a JSPM override to customize the exports field for this package.

Readme

@aiiware/aii

AI-powered CLI agent with autonomous tool execution, multi-agent delegation, and multi-channel access. An intelligent assistant that can read, write, search, and execute commands in your terminal — or through Telegram.

Installation

npm install -g @aiiware/aii

Requirements

Node.js 18+
Aii Server running (default: localhost:26169)

Quick Start

# Start interactive agent mode
aii

# Run a single task
aii "explain what this project does"

# Pipe content for processing
cat error.log | aii "summarize the errors"

# Auto-approve all tool executions
aii --auto

Agent Mode (Default)

Aii runs as an intelligent agent that can autonomously execute tools to complete tasks:

aii                    # Start interactive session
aii "your task"        # Run single task and exit

Available Tools

Tool	Description
Read	Read file contents
Write	Create or overwrite files
Edit	Make targeted edits to files
Glob	Find files by pattern
Grep	Search code with regex
Bash	Execute shell commands
WebSearch	Search the web
WebFetch	Fetch and parse web pages
Task	Delegate work to specialized subagents

Interactive Commands

/help              Show available commands
/clear             Clear conversation history
/new               Start a fresh conversation
/history           Show conversation summary
/model <model>     Switch LLM model
/provider <name>   Switch LLM provider
/context           View context window usage
/stats             Show session statistics (tokens, cost)
/init              Generate AGENTS.md for your project
/memory            View loaded project instructions
/hooks             View configured hooks
/skills            List available skills
/commit            Generate commit message and commit
/loop              Start autonomous agent loop
/mcp               Manage MCP servers
/agents            List available subagents
/tasks             Inspect subagent task executions
/diff              Show recent file changes
/undo              Undo the last file change
/mode              Show or switch operation mode (normal/auto/plan)
/exit              Exit the session

Keyboard Shortcuts

Key	Action
`Enter`	Submit input
`Esc`	Cancel current request
`Ctrl+C`	Exit immediately
`Ctrl+D`	Graceful exit with session summary
`Ctrl+L`	Clear conversation
`Up/Down`	Navigate input history
`Tab`	Select from command menu

Multi-Agent Subagents (v0.10)

Delegate complex tasks to specialized subagents that run in parallel within the same session:

> analyze my codebase architecture and review the auth module

  ● Running 2 agents...
    ├─ Explore (Analyze codebase architecture) · 8 tool uses
    │    └ Running (5s)
    └─ Review (Review auth module) · 3 tool uses
         └ Running (2s)

Built-in Subagents

Subagent	Purpose
explore	Fast codebase exploration (read-only)
plan	Architecture and implementation design
review	Code quality analysis

Custom Subagents

Define your own subagents in .aii/agents/ with custom system prompts, model tiers, and tool access levels.

Task Inspection

/tasks              # List all task executions with IDs
/tasks <id>         # Show full transcript for a specific task

Telegram Bot Channel (v0.10)

Access your AI agent remotely through Telegram — works anywhere, no port forwarding needed:

# Pair a Telegram bot
aii telegram pair mybot <bot-token>

# Start the bot listener
aii telegram start mybot

# Manage multiple bots
aii telegram list
aii telegram status mybot
aii telegram unpair mybot

Bot commands in Telegram: /start, /help, /reset, /status, /compact, /model

MCP Integration (v0.10)

Extend the agent with external tools via Model Context Protocol:

/mcp catalog                    # Browse available servers
/mcp install github             # One-command install from catalog
/mcp add myserver cmd args...   # Add custom server
/mcp list                       # Show configured servers
/mcp remove myserver            # Remove a server

Once installed, MCP tools are available to the agent automatically.

Autonomous Loops (v0.7)

Run the agent in a continuous loop for long-running autonomous tasks:

# Start a loop from CLI
aii loop --verify "npm test" --duration 5m

# Or start within a session
/loop --max 10 --promise "All tests passing"

Options:

Option	Description
`--max <N>`	Maximum iterations
`--verify "<cmd>"`	Run verification command after each step
`--promise "<text>"`	Stop when output contains this text
`--budget <$N>`	Cost limit
`--duration <time>`	Time limit (e.g., `30m`, `2h`)

LLM Providers

Aii supports multiple LLM providers. Switch anytime with /model or /provider:

Provider	Models	Setup
Anthropic	Claude Sonnet, Opus, Haiku	API key
OpenAI	GPT-4o, GPT-4, o1, o3	API key
Google	Gemini Pro, Flash	API key
DeepSeek	DeepSeek Chat, Coder	API key
Ollama	Llama, Mistral, and more	Local, no API key
OpenRouter	300+ models via single API key	API key

# Switch provider and model mid-session
/model claude-sonnet-4
/model ollama/llama3.1:8b
/provider openrouter

The provider is auto-detected from model names (e.g., gpt-4o routes to OpenAI).

Context Window Management (v0.3)

Prevents "token limit exceeded" errors with intelligent auto-summarization:

Status bar shows real-time context usage with color-coded alerts
Auto-summarization triggers when approaching the limit, preserving recent context
Dynamic thresholds adjust per model's output capacity

/context            # View detailed usage breakdown
/context clear      # Manually trigger summarization

Project Instructions (v0.5)

Teach the agent your project's conventions with AGENTS.md:

/init                   # Auto-generate by analyzing your project
/init --with-hooks      # Also generate hooks.json template
/memory                 # View loaded instructions
/memory reload          # Hot-reload without restart

The agent discovers AII.md or AGENTS.md in your project root, plus ~/.aii/AGENTS.md for global preferences.

Hooks (v0.5)

Run custom scripts before/after tool execution:

/init --with-hooks      # Generate hooks template
/hooks                  # View configured hooks
/hooks reload           # Hot-reload without restart

Hook types: PreToolUse, PostToolUse, Stop, Iteration

Skills (v0.6)

Extend the agent with reusable skill packages:

Natural language — the agent auto-detects the right skill:

> commit my changes
> review the latest PR

Slash commands (explicit):

/commit              # Generate commit message and execute
/review-pr 123       # Review a pull request

Managing skills:

/skills                                        # List all skills
/skills install aiiware/skills/code-review     # Install from GitHub
/skills install aiiware/skills/refactor --user # Install globally
/skills uninstall code-review                  # Remove a skill
/skills reload                                 # Re-discover skills

Creating skills — add SKILL.md files in .aii/skills/:

---
name: my-skill
description: Does something useful
allowed-tools: Bash Read Grep
---

Instructions for the agent...
Use $ARGUMENTS for user input.

Configuration

aii config show                # Show current configuration
aii config validate            # Validate configuration
aii config set <key> <value>   # Set a configuration value
aii config model <model>       # Change LLM model
aii config provider <provider> # Change LLM provider

Configuration is stored in ~/.aii/:

config.yaml — Main configuration:

llm:
  provider: anthropic
  model: claude-sonnet-4
  temperature: 0.7

api:
  url: http://localhost:26169

secrets.yaml — API keys:

anthropic_api_key: sk-ant-...
openai_api_key: sk-...
deepseek_api_key: sk-...

Or use environment variables: ANTHROPIC_API_KEY, OPENAI_API_KEY, DEEPSEEK_API_KEY

CLI Options

Option	Description
`--auto`	Auto-approve all tool executions
`--model <model>`	Override LLM model
`--provider <provider>`	Override LLM provider
`--host <host:port>`	API server host
`-c, --continue-chat <id>`	Continue a previous session
`-n, --new-chat`	Force a new session
`--offline`	Disable web and MCP tools
`--clean`	Clean output mode
`--standard`	Standard output mode (default)
`--thinking`	Show full reasoning
`--no-streaming`	Disable streaming output
`-q, --quiet`	Minimal output
`-y, --yes`	Auto-confirm shell commands

System Commands

aii doctor         # System health check
aii stats          # Usage analytics (models, cost)
aii --help         # Show help
aii --version      # Show version

Features

Autonomous Agent — Executes multi-step tasks with tool use
Multi-Agent Delegation — Parallel subagents for complex tasks
Telegram Access — Use your agent remotely via Telegram bots
MCP Integration — Extend with external tools via Model Context Protocol
Autonomous Loops — Continuous agent execution with verification and budgets
Multiple Providers — Claude, GPT, Gemini, DeepSeek, Ollama, OpenRouter
Real-time Streaming — Token-by-token response display
Context Management — Auto-summarization prevents token limit errors
Project Instructions — Teach the agent your conventions with AGENTS.md
Skills System — Extend with reusable skills, invokable via commands or natural language
Hooks — Custom scripts for validation, logging, and approval gates
Session Persistence — Continue previous conversations with --continue-chat

JSPM

@aiiware/aii

Package Exports

Readme

@aiiware/aii

Installation

Requirements

Quick Start

Agent Mode (Default)

Available Tools

Interactive Commands

Keyboard Shortcuts

Multi-Agent Subagents (v0.10)

Built-in Subagents

Custom Subagents

Task Inspection

Telegram Bot Channel (v0.10)

MCP Integration (v0.10)

Autonomous Loops (v0.7)

LLM Providers

Context Window Management (v0.3)

Project Instructions (v0.5)

Hooks (v0.5)

Skills (v0.6)

Configuration

CLI Options

System Commands

Features

Links

License