janito - a development agent with function calling, MCP and skills

Console vs Web

📖 Full documentation available at https://joaompinto.github.io/janito/

⚠️ Disclaimer: The code on this repo has been developed mostly using AI, more specifically the Qwen models.

Features

🔧 Function Calling - Built-in tools for file operations, web search, and more
🔌 Plugins - Gmail, code search, OneDrive, and more via the plugin system
🔌 MCP Support - Connect to Model Context Protocol servers
🧩 Skills - Install and use task-specific skills from GitHub
🌐 Web UI (Alpha) - Chat through a browser instead of the terminal with --web
📊 Real-time Progress - Watch tool execution progress as it happens
🚀 Easy Setup - Interactive configuration with --config or quick setup with --set flags
🔗 OpenAI-Compatible & Native APIs - Works with any OpenAI-compatible endpoint (OpenAI, LM Studio, Ollama, custom) and native SDKs (Anthropic, DashScope)

NOTE: Janito should be used on development systems on which data loss is tolerable and there are no critical secrets

Quick Start

# Install
pip install janito

# Or, with uv (recommended)
uv tool install janito

# Configure interactively
janito --config

# Or set options directly (two steps: config, then API key)
janito --set provider=openai --set model=gpt-4
janito --set-api-key="sk-your-key" --provider openai

# Start chatting
janito "Hello!"

Installation

From PyPI

pip install janito

Or, with uv:

uv tool install janito

For development setup, see README_DEV.md.

Configuration

Interactive Setup

janito --config

You'll be prompted for:

Provider - openai or custom
API Key - Masked for security
Model - e.g., gpt-4, gpt-3.5-turbo
Max Output Tokens - Maximum output tokens (default: 65536)
Max Input Tokens - Maximum input tokens / context window (default: 128000)

Quick Configuration with `--set`

Set options directly from the command line:

# Single key-value
janito --set model=gpt-4

You can also use --get, --unset, and --set-secret with multiple values.

View Configuration

janito --show-config

Costs per provider

Provider	Last Model	Price
Alibaba	qwen3.8-max	6$/month (Lite)
Moonshot	kimi-k3	$15/month (Moderato)
z.ai	glm-5.3	$12.6/month (Lite)
openai	gpt-5.6	Pay as you go

For custom endpoints (base-url), see README_custom.md.

Usage

Single Prompt

janito "What is the capital of France?"

Pipe Input

echo "Tell me a joke" | janito

Interactive Chat

janito

Commands in chat mode:

exit / quit - End session
clear - Clear conversation and start a new one
Ctrl+D / Ctrl+Z - Exit

Web UI (Alpha)

⚠️ Alpha — The --web mode is currently in alpha. It works, but the interface, CLI flags, and API may change between releases.

Janito can serve a browser-based chat interface instead of the terminal. This requires optional dependencies that are not part of the core install:

# Install with the web extra
pip install janito[web]
# or
uv tool install janito[web]

Then start the server (opens your browser automatically):

janito --web

# All normal Janito flags still apply, plus web-specific ones:
janito --web -r -w --web-port 9090    # privileges + web, custom port
janito --web --no-web-open                      # don't auto-open the browser

The server binds to 127.0.0.1 by default (localhost only). For full details (features, security, API endpoints, architecture), see the Web UI documentation.

System Prompt Options

Control how the system prompt is handled:

# Use default system prompt (with tools enabled)
janito "What can you do?"

# Disable system prompt entirely (no tools at all)
janito -Z "Simple prompt without system context"

# Override with a custom system prompt (tools stay enabled)
janito -S "You are a concise coding assistant" "Explain recursion"

# Keep the system prompt but skip tool loading (skill tools stay enabled)
janito --no-tools "Explain recursion"

Flag	Description
`-Z`, `--no-system-prompt`	Skip system prompt and disable tools
`-S`, `--system-prompt`	Custom system prompt
`--no-tools`	Do not load tools (skill tools stay enabled)

Note: When using -Z, built-in tools (file operations, MCP) are disabled. Use the default mode or -S flag when you need tool access.

Note: --no-tools disables loading of all non-skill tools (file operations, MCP) while keeping the skill tools (load_skill, read_skill_resource) available, so the model can still load installed skills on demand.

Logging

janito --log=info "Your prompt"      # Info level
janito --log=debug "Your prompt"     # Debug level
janito --log=info,debug "Your prompt" # Multiple levels

Examples

OpenAI

# Step 1: Set provider and model
janito --set provider=openai --set model=gpt-4
# Step 2: Store API key
janito --set-api-key="sk-your-key" --provider openai

# Then run any prompt
janito "Explain quantum computing"

Alibaba (Qwen)

# Step 1: Set provider and model
janito --set provider=alibaba --set model=qwen-plus
# Step 2: Store API key
janito --set-api-key="your-dashscope-api-key" --provider alibaba

# Then run any prompt
janito "Explain quantum computing"

Custom Endpoint

janito --set provider=custom --set base-url=http://localhost:8000/v1

Built-in Tools

janito includes tools for common tasks:

File Operations

# List files
python -m janito.tools.files.list_files . --recursive --pattern "*.py"

# Read file
python -m janito.tools.files.read_file README.md --max-lines 20

MCP Tools

Connect to MCP servers using the /mcp command inside the interactive shell:

# Add a stdio-based MCP server
/mcp add myserver stdio python -m mcp.server

# Add an HTTP-based MCP server
/mcp add remote http https://api.example.com/mcp

# List configured servers
/mcp list

For full MCP documentation, see README_MCP.md.

Tool Progress Reporting

Tools report progress in real-time:

🔄 Reading files...
📊 Processing: 50/100 files
✅ Completed: 100 files (2.3MB)

Progress messages go to stderr so they don't interfere with tool output.

Error Handling

Exit Code	Meaning
`0`	Success
`1`	Configuration or runtime error
`130`	User cancelled (Ctrl+C or Enter)

Dependencies

Python 3.10+
openai>=1.0.0
rich>=10.0.0
prompt-toolkit>=3.0.0
requests>=2.28.0 (for MCP support)
pathspec>=0.11.0 (for .gitignore-aware file listing)

License

MIT License

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

janito-4.31.0.tar.gz (905.5 kB view details)

Uploaded Aug 25, 2026 Source

Built Distribution

If you're not sure about the file name format, learn more about wheel file names.

The dropdown lists show the available interpreters, ABIs, and platforms. Enable javascript to be able to filter the list of wheel files.

janito-4.31.0-py3-none-any.whl (549.8 kB view details)

Uploaded Aug 25, 2026 Python 3

File details

Details for the file janito-4.31.0.tar.gz.

File metadata

Download URL: janito-4.31.0.tar.gz
Upload date: Aug 25, 2026
Size: 905.5 kB
Tags: Source
Uploaded using Trusted Publishing? Yes
Uploaded via: twine/7.0.0 CPython/3.13.14

File hashes

Hashes for janito-4.31.0.tar.gz
Algorithm	Hash digest
SHA256	`b996088530026bd8cb350ca1967f6e56666ac9a843c0fb7da2c0f35412354e8e`
MD5	`e0fbe31e8daceb2c2460aefc688137da`
BLAKE2b-256	`8739a33385aec8dcfafdd44953c216185cf9beaaa666ef2b9b24527cc3719bf3`

See more details on using hashes here.

Provenance

The following attestation bundles were made for janito-4.31.0.tar.gz:

Publisher: release.yaml on joaompinto/janito

Attestations: Values shown here reflect the state when the release was signed and may no longer be current.

Statement:
- Statement type: https://in-toto.io/Statement/v1
- Predicate type: https://docs.pypi.org/attestations/publish/v1
- Subject name: janito-4.31.0.tar.gz
- Subject digest: b996088530026bd8cb350ca1967f6e56666ac9a843c0fb7da2c0f35412354e8e
- Sigstore transparency entry: 2583852849
- Sigstore integration time: Aug 25, 2026
Source repository:
- Permalink: joaompinto/janito@e36dda4dac95749d106a8d672f7c0dc317f715d6
- Branch / Tag: refs/tags/v4.31.0
- Owner: https://github.com/joaompinto
- Access: public
Publication detail:
- Token Issuer: https://token.actions.githubusercontent.com
- Runner Environment: github-hosted
- Publication workflow: release.yaml@e36dda4dac95749d106a8d672f7c0dc317f715d6
- Trigger Event: push