tiaga 0.2.30

A CLI tool

Tiaga

Author: Seems Kushwaha (seemsyt)

Tiaga is a powerful terminal-based AI assistant designed for coding and tool-assisted workflows in your local workspace. It provides an interactive TUI interface with comprehensive tool support, session persistence, and advanced configuration options.

Features

• Interactive TUI Interface: Rich terminal UI with real-time streaming responses
• Built-in Tools: File operations (read, write, edit), shell execution, directory listing, search (grep, glob), web search, web fetch, YouTube transcript, todo management, memory storage
• MCP Support: Model Context Protocol server integration for extended capabilities
• Session Persistence: Save, resume, and checkpoint conversations
• Hooks System: Execute custom commands/scripts at various trigger points
• Approval Policies: Configurable approval for tool execution (on_request, on_failure, auto, auto_edit, never, yolo)
• Smart Context Management: Automatic context compression when window limit is approached
• Loop Detection: Prevents repetitive tool calls
• Multi-provider Support: Compatible with OpenAI-style APIs (OpenRouter, etc.)
• Config Scopes: User-level and project-level configuration
• Shell Environment Control: Configurable environment variables and exclusion patterns
• Developer Instructions: Custom system prompts at user and developer levels

Installation

Using pip:

pip install tiaga or pipx install tiaga

Or from source:

git clone <repository>
cd tiaga
pip install -e .

Configuration

Environment Variables

Set these before running:

API_KEY=your_api_key_here
BASE_URL=https://openrouter.ai/api/v1

Optional environment variable:

• TIAGA_DEBUG=1 - Enable debug logging

Configuration File

Tiaga supports TOML configuration at two levels:

1. User config: ~/.local/share/seems-tiaga/config.toml (Linux) or equivalent
2. Project config: <project>/.seems-tiaga/config.toml (overrides user config)

Example config.toml:

[model]
name = "stepfun/step-3.5-flash:free"
temperature = 1.0

[approval]
policy = "on_request"  # or "on_failure", "auto", "auto_edit", "never", "yolo"

[cwd]
path = "/path/to/working/directory"

[shell_environment]
excludes_patterns = ["*KEY*", "*TOKEN*", "*SECRET*"]
set_vars = { "CUSTOM_VAR" = "value" }

[mcp_servers.example]
command = "npx"
args = ["-y", "@modelcontextprotocol/server-filesystem", "/path/to/files"]

[hooks.pre_agent]
trigger = "before_agent"
command = "python3 hooks/before_agent.py"
timeout_sec = 10
enabled = true

[max_turns] = 100
[max_output_tokens] = 50000
[debug] = true

Usage

Interactive Mode

tiaga

Or with specific working directory:

tiaga --cwd /path/to/project

Project Graph Commands

Generate and save a dependency graph under ./project_graph/:

tiaga graph --format tree
tiaga graph --format md
tiaga graph --format json

Explain a file's role and relationships (writes a report file under ./project_graph/):

tiaga explain tiaga/agent/agent.py

Get an impact report for changes to a file (writes a report file under ./project_graph/):

tiaga impact tiaga/config/config.py

Single Prompt Mode

tiaga "Write a Python function to calculate fibonacci numbers"

Command-line Options

• --cwd, -c: Set the working directory (must exist)

CLI Commands

During interactive sessions, use these slash commands:

Navigation & Control

• /help - Show help
• /exit, /quit, /q - Exit Tiaga
• /clear - Clear conversation history and loop detector state

Configuration

• /config - Show current configuration
• /config model <name> - Change model
• /config base_url <url> - Change base URL
• /config api_key <key> - Change API key
• /config approval <policy> - Change approval policy (on_request, on_failure, auto, auto_edit, never, yolo)

Model Management

• /model - Show current model and token usage
• /model <name> - Switch to a different model

Tools

• /tools - List all available tools
• /mcp - List MCP servers

Session Persistence

• /save - Save current session
• /sessions - List all saved sessions
• /resume <session_id> - Resume a saved session
• /checkpoint - Create a checkpoint of current session
• /checkpoints <session_id> - List checkpoints for a session
• /restore <checkpoint_id> - Restore from a checkpoint

Debug & Stats

• /stats - Show session statistics (turns, tool calls, token usage, etc.)

CLI Commands Reference

Command	Description
/help	Show this help message
/exit / /quit / /q	Exit the application
/clear	Clear conversation history
/config	Display current configuration
/config model <name>	Set LLM model
/config base_url <url>	Set API base URL
/config api_key <key>	Set API key
/config approval <policy>	Set approval policy
/model	Show current model and usage
/model <name>	Change model
/tools	List available tools
/mcp	List MCP servers
/save	Save session to disk
/sessions	List saved sessions
/resume <id>	Resume a saved session
/checkpoint	Create checkpoint
/checkpoints <session_id>	List checkpoints
/restore <checkpoint_id>	Restore from checkpoint
/stats	Show session statistics
/approval	Show/change approval policy

Approval Policies

Control when tool executions require user approval:

• on_request (default) - Ask for approval for each tool call
• on_failure - Auto-approve, ask only if tool fails
• auto - Auto-approve all tools (use with caution)
• auto_edit - Auto-approve only file edit tools
• never - Never ask for approval (not recommended)
• yolo - No approval, no safety checks (dangerous!)

Set with: /config approval <policy> or in config file.

Built-in Tools

Tiaga includes these tools by default:

Tool	Description
readfile	Read file contents
writefile	Write content to a file
editfile	Edit file with search/replace
shell	Execute shell commands
listdir	List directory contents
grep	Search text in files
glob	Find files by pattern
websearch	Search the web (DuckDuckGo)
webfetch	Fetch URL content
youtube_scrapping	Get YouTube video transcript
todo	Manage task list
memory	Store/retrieve persistent memory
project_graph_refresh	Rebuild and refresh graphs under project_graph/ (for long sessions)

Project Layout

tiaga/
├── __init__.py           # Package init with warnings setup
├── main.py               # CLI entry point, command handling
├── hello.py              # Simple hello function
├── dummy_script.py       # Example script
│
├── agent/                # Agent orchestration
│   ├── agent.py          # Core Agent class with run loop
│   ├── events.py         # AgentEvent types and definitions
│   ├── session.py        # Session management, tool registry
│   └── persistence.py    # Session checkpoint/save/load
│   └── planner.py
|
|
├── client/               # LLM client layer
│   ├── llm_client.py     # OpenAI-compatible client
│   └── response.py       # Response parsing, token usage
│
├── config/               # Configuration management
│   ├── config.py         # Config model (pydantic)
│   └── loader.py         # Config loading & merging
│
├── context/              # Conversation context
│   ├── manager.py        # Message history, token counting
│   ├── prompts.py        # System prompts, loop breaker
│   ├── text.py           # Text utilities
│   └── compaction.py     # Context compression
│
├── hooks/                # Hook system
│   └── hook_system.py    # Hook trigger management
│
├── safty/                # Safety/approval (note: typo in dir name)
│   └── approval.py       # Approval manager, callbacks
│
├── scripts/              # Utility scripts
│   └── test_tool.py      # Tool testing script
│
├── tools_manager/        # Tool infrastructure
│   ├── base.py           # Base Tool class, ToolResult
│   ├── registry.py       # ToolRegistry, tool discovery
│   ├── subagent.py       # Sub-agent tool wrapper
│   ├── discovery.py      # Tool discovery utilities
│   ├── buildin/          # Built-in tools (12 tools)
│   │   ├── readfile.py
│   │   ├── writefile.py
│   │   ├── editfile.py
│   │   ├── shell.py
│   │   ├── listdir.py
│   │   ├── grep.py
│   │   ├── glob.py
│   │   ├── websearch.py
│   │   ├── webfetch.py
│   │   ├── youtube_scrapping.py
│   │   ├── todo.py
│   │   └── memory.py
│   ├── mcp/              # MCP server integration
│   │   ├── manager.py
│   │   ├── transport.py
│   │   └── types.py
│   └── __init__.py
│
|
├──tracing/
|       ├──trace.py
├── ui/                   # User interface
│   ├── render.py         # TUI rendering, panels, streaming
│   └── __init__.py
│
└── utils/                # Utilities
    ├── config_setter.py  # Empty module
    ├── erors.py          # Custom exceptions
    ├── path.py           # Path utilities
    └── __init__.py

# Root level files
├── pyproject.toml        # Project metadata, dependencies
├── uv.lock               # Dependency lock file
├── README.md             # This file
├── test.py               # Simple test
└── dist/                 # Build artifacts

Development

Quick Check

Verify the package compiles:

python -m compileall tiaga

Running Tests

python test.py

Building

pip install build
python -m build

How It Works

1. Initialization: Loads config from user and project scopes, sets up LLM client
2. Session Creation: Creates a session with context manager, tool registry, and MCP manager
3. Message Processing: User input is added to context, agent enters loop:
   • Get tool schemas from registry
   • Call LLM with context
   • Stream response back to user
   • Execute any tool calls (with approval if required)
   • Add tool results to context
   • Check for context compression need
   • Repeat up to max_turns
4. Event Streaming: Events (tool calls, text deltas, completions, errors) are yielded and rendered in TUI
5. Persistence: Sessions can be saved and resumed at any point

Extensibility

Adding Custom Tools

Create a Python file and register tools using the @tool decorator:

from tiaga.tools_manager.base import tool, Tool
from pydantic import BaseModel

class MyInput(BaseModel):
    param: str

@tool
def my_tool(input: MyInput) -> str:
    """Tool description"""
    return f"Result for {input.param}"

Adding MCP Servers

Configure in config.toml:

[mcp_servers.my_server]
command = "path/to/mcp/server"
args = ["--flag", "value"]

License

MIT