LansCoder

A local coding agent you can read end to end.

What is it?

LansCoder is a locally-runnable Python coding agent. It understands your codebase, edits files, and runs shell commands — like Claude Code or Aider. But its core design goal isn't feature volume; it's understandability.

~28,000 lines of Python, clean module boundaries, solid test coverage. Real enough to use daily, small enough to read from end to end.

96.38% reward pass@1 on the Harbor Aider Polyglot benchmark.

Highlights

• Small, so you can read it — ~28k lines of Python, not 570k lines of TypeScript. Every module's job is obvious.
• Built to learn from — strict layering, clear dependency rules. Great for studying how coding agents work, for hacking on, or for interview prep.
• Actually usable — 29 built-in tools, multi-provider support, MCP integration, session persistence, context compression. Not a toy.
• Preview before you write — syntax-highlighted diffs shown before every file change, even in high-permission mode; confirmation via strict 1/2/3 input, with reject: <feedback> for write review.

Quick start

Choose your preferred install method:

Shell (one-liner, all platforms):

curl -sSL https://raw.githubusercontent.com/Lanstzz/LansCoder/main/install.sh | bash

pipx (all platforms):

pipx install lanscoder

After installation, run lanscoder config init to generate a config file, then edit it with your API key:

• macOS / Linux: ~/.config/lanscoder/config.toml
• Windows: C:\Users\<username>\.config\lanscoder\config.toml

Example config:

default_model = "deepseek/deepseek-v4-flash"

[providers.deepseek]
type = "openai-compatible"
base_url = "https://api.deepseek.com"
api_key = "sk-xxx"

[models."deepseek/deepseek-v4-flash"]
label = "DeepSeek V4 Flash"
context_window = 1000000

[permissions]
mode = "ask"

[ui]
theme = "default"

Then launch from your project directory:

lanscoder

Development setup:

python -m venv .venv
.venv/bin/python -m pip install -e ".[dev]"
.venv/bin/python -m pytest

Features at a glance

Architecture

lanscoder/
├── app/           Textual TUI
├── agent/         Agent loop, tool execution, permission resume
├── providers/     Model provider adapters
├── tools/         Tool registration and execution (29 built-in tools)
├── permissions/   Policy, grants, and permission coordinator
├── context/       Event log and context management
├── session/       Session lifecycle
├── planning/      Task plan service and projection
├── subagent/      Background subagent types
├── input/         Attachments and clipboard
├── mcp/           MCP protocol integration
├── memory/        Cross-session persistent memory
├── skills/        Local skill discovery and loading
├── config/        TOML configuration
└── utils/         Shared utilities

Who is it for?

• Developers who want to deeply understand how coding agents work
• People looking to extend or modify a Python-based coding agent
• Anyone who needs a project they can explain architecturally for interviews or portfolios
• AI enthusiasts who want to experiment with different models locally

Contributing

Issues and PRs are welcome. Tests cover most core modules — please make sure they pass before submitting.

License

MIT