quenda 0.3.1

A lightweight, layered Agent framework for Python

What's New

2026-06 (v0.3.0)

• Skills Framework — Composable capability packages with instructions, resources, and tools. Discover, activate, and use skills on demand. Learn more →
• Context Compression — Automatic summarization when context grows large, with configurable policies and manual /compress command.
• Interaction Requests — LLM can ask users for choices, confirmations, and input via structured interaction protocol.
• Custom Tool Extensions — Agent-local tools via extensions/tools/*.py with config.yaml declaration.
• Policy System — Runtime policies for termination, tool selection, and result processing.
• Multimodal Support — Image input support with resource activation.
• 26 Model Providers — OpenAI, Anthropic, DeepSeek, DashScope, Moonshot, Ollama, and 20 more.

Quenda Code

Quenda Code is an AI coding agent that runs in your terminal. It reads your codebase, writes code, runs commands, and helps you ship.

📖 Quenda Code Documentation →

# Install
pip install quenda quenda-code

# Start interactive session
quenda code

# One-shot task
quenda code "Add error handling to the API client"

Features

• Code-aware — Reads and understands your codebase structure
• Workspace-scoped — All file operations stay within your project directory
• Safe execution — Shell commands filtered, Python code sandboxed
• Session persistence — Conversations saved and resumable
• Interactive mode switching — Switch between code, architect, and chat modes
• Skills on demand — Activate capability packages as needed

REPL Commands

Example Session

> read the main entry point and explain how it works

I'll read the main entry point...

[Reads src/quenda/cli.py]

The entry point is `cli.py:main()`. It defines two subcommands:
- `quenda run --agent <path>` — Run a custom agent from AGENT.md
- `quenda code` — Run Quenda Code Agent (built-in)

Each command supports one-shot mode (with a message) or REPL mode (without).

> add a --version flag to the CLI

I'll add a `--version` flag to the argument parser...

[Applies patch to cli.py]

Done. Added `--version` flag that prints the version and exits.

> run the tests

[Runs pytest]

All 42 tests passed.

Quenda SDK

Build agents in Python with a minimal API:

from quenda import Agent, tool
from quenda.providers import get_provider_registry
from quenda.tools import get_core_tools
import asyncio

@tool
def calculate(expression: str) -> float:
    """Safely evaluate a math expression."""
    import ast
    node = ast.parse(expression, mode='eval')
    return eval(compile(node, '<string>', 'eval'), {"__builtins__": {}}, {})

model = get_provider_registry().get_model("deepseek", "deepseek-v4-flash")

agent = Agent(
    name="assistant",
    system_prompt="You are a helpful assistant.",
    tools=[calculate, *get_core_tools(".")],
    model=model,
)

async def main():
    session = agent.open_session()
    result = await session.send("What is 15% of 847?")
    print(result)

asyncio.run(main())

📖 SDK Tutorials — 8 chapters covering agents, tools, providers, sessions, and events.

Installation

# Quenda Code — AI coding assistant (CLI)
pip install quenda quenda-code

# Quenda SDK — Build agents in Python
pip install quenda

Requires Python 3.12+. Zero required runtime dependencies.

Features

• Minimal API. Agent, Session, @tool, and you're done.
• 26 model providers. OpenAI, Anthropic, DeepSeek, DashScope, and more — one registry, one API.
• 9 core tools. Filesystem, shell, Python sandbox, and user interaction — all workspace-scoped.
• Skills framework. Composable capability packages with instructions and resources.
• Security by code. SSRF protection, command filtering, import restrictions, workspace isolation.
• Observable by default. Every run emits structured events for streaming and debugging.
• Context compression. Automatic summarization when context grows large.

Model Providers

Quenda ships with 26 built-in providers covering 300+ models:

Full provider list →

Add a custom provider in 5 lines:

from quenda.providers import ProviderSpec, ModelSpec, get_provider_registry

registry = get_provider_registry()
registry.register(ProviderSpec(
    id="my-provider",
    name="My Provider",
    base_url="https://api.example.com/v1",
    api="openai-completions",
    api_key="${MY_API_KEY}",
    models=(ModelSpec(id="my-model", name="My Model", tool_calling=True),),
))

Built-in Tools

get_core_tools(workspace) returns 10 essential framework tools:

Full tool reference →

Architecture

Interface → Host → Runtime → Kernel

Each layer depends only on the layer inside it. The Kernel is fully testable with fake models — no network required.

Documentation

Contributing

Quenda is intentionally small. Before making a change, read CLAUDE.md and the ADR records.

1. Identify which architectural layer owns the change.
2. Prefer the smallest complete change; add tests with behavior changes.
3. Do not cross established layer boundaries.

pip install -e ".[dev]"   # editable install with dev tooling
pytest                    # run tests
ruff check src/quenda       # lint

License

MIT