Ainator

Ainator is a CLI/API framework for AI agents built on the agno framework. It provides interactive chat, persistent sessions, knowledge bases (RAG), and built-in tools for shell execution, file operations, and web search.

Features

• Interactive Chat: Persistent sessions with compression and summarization
• Knowledge Bases (RAG): Embed code repositories (AST-aware chunking), websites, and directories
• Built-in Tools: Shell execution (with safety controls), file operations, DuckDuckGo search
• Session Management: List, switch, show, and compress sessions
• Learning & Memory: User profile, memory, and session context
• Cultural Knowledge: Context-aware cultural knowledge integration
• KeyboardInterrupt Handling: Proper handling of interruption signals
• Skills System: Extensible via .ainator/skills/ directory
• MCP Support: Local, remote, any MCP is configurable
• Web Server: FastAPI server with Swagger documentation
• Model Support: VLLM, DeepSeek, OpenAI, Anthropic...

Installation

pip install ainator

Quick Start

# Run a one-off prompt (creates new session)
ainator hi "What is the capital of France?"

# Continue in existing session
ainator but "Tell me more about that"

# Or start interactive chat
ainator chat

# Or Start the web server (localhost:7777/docs)
ainator server
# Then consume it through https://github.com/agno-agi/agent-ui

Commands

Chat & Prompt

• ainator hi "your prompt" - One-off prompt execution (creates new session)
• ainator but "your prompt" - Continue in existing session
• ainator chat - Interactive chat session

Test-Driven Development & Fixing

• ainator fix "python my_script.py or any command or pytest" - Run until rc == 0

Session Management

• ainator sessions - List all sessions
• ainator switch <id> - Switch to a session
• ainator show - Show messages in current session

Skills Management

• ainator skills - List available skills

RAG Management

• ainator rag code add ./path/to/repo python - Add code repository
• ainator rag website add <url> - Add website content
• ainator rag search <knowledge-name> "query" - Search knowledge base
• ainator rag remove <knowledge-name> - Remove knowledge base
• ainator rag - Show RAG status

Server

• ainator server - Start FastAPI web server (port 7777)

Configuration

Configuration file is auto-managed, but you can still hack it if you want to tweak parameters: .ainator/config.yml

Example Configuration

agent:
    add_culture_to_context: true
    add_history_to_context: true
    compress_tool_results: true
    enable_session_summaries: true
    reasoning: true
    reasoning_max_steps: 5
    reasoning_min_steps: 2
    store_events: true
    update_cultural_knowledge: true
    update_memory_on_run: true
compression_manager: agno.compression.manager:CompressionManager compress_token_limit=60000 compress_tool_results=True
compression_model: agno.models.vllm:VLLM id=coder
db: ainator.db.sqlite:AsyncSqliteDb db_file=/home/jpic/src/ainator/.ainator/agno.db
extra_skills: []
knowledge:
    agno:
        path: /home/jpic/src/agno/libs/agno/agno
        plugin: code
learning_machine: agno.learn:LearningMachine user_profile=False user_memory=False session_context=True entity_memory=False learned_knowledge=True
model: agno.models.vllm:VLLM id=coder
session_id: bf51cd29-c374-4074-b0eb-faa54f2a6e70
session_name: Fixing Configuration Tests with Temporary Directories
tools:
- agno.tools.file:FileTools base_dir=/home/jpic/src/ainator
- agno.tools.shell:ShellTools
- agno.tools.duckduckgo:DuckDuckGoTools
mcp:
- url: https://docs.agno.com/mcp
- commad: npx -y @modelcontextprotocol/server-google-maps

Model Configuration Examples

# VLLM
model: agno.models.vllm:VLLM id=coder

# DeepSeek
model: agno.models.deepseek:DeepSeek id=deepseek-chat

Environment Variables

• AINATOR_MODEL - Override the model configuration

Architecture

Core Framework

• agno: Core AI agent framework (agents, models, tools, SQLite database)
• cli2: Command-line interface framework

Knowledge Plugins

• code: Repository embedding with AST-aware chunking
  • Uses llama-index CodeSplitter and HierarchicalNodeParser
  • BAAI/bge-small-en-v1.5 embeddings
  • LanceDB vector storage
• site: Website scraping with Parsel
• generic: Directory embedding descriptive file names is required

Skills System

Extensible via .ainator/skills/ directory, just create a skill in there or clone a skills repo such as:

git clone https://github.com/anthropics/skills .ainator/skills/anthropics

Renderer

Streaming Markdown output

Server

FastAPI server via AgentOS, consumable with AgentUI or the AgentOS client, or plain HTTP even.

Tools

• Enhanced shell execution with chunked output
• File operations
• DuckDuckGo search
• Custom knowledge tools (RAGs)

Compression

Automatic

Learning

• User profile management
• Memory system
• Session context tracking

Development

Installation for Development

pip install -e '.'

Testing

pytest tests/

Version

Version 0.0.3 (defined in src/ainator/__about__.py)

License

MIT © 2026-present jpic

Links

• PyPI
• Source Code
• Issues
• Agno Framework
• cli2