wow-agent
v0.6.0 · MIT · Terminal Native · ARM64

The AI Coding Partner
in Your Terminal

Give it a goal — wow-agent plans, executes tools, runs your tests, and rolls back on failure. Any model. Your data never leaves the machine.

Get Started → GitHub 
git clone https://github.com/uuzjw/wow-agent && cd wow-agent && uv sync && uv run wow

One interface, every major model

DeepSeekQwenKimiGLMOpenAIGeminiGroqOpenRouterOllamaLM Studio

Choose your path

Just want to get things done, or build on top of it? We wrote docs for both of you.

🚀

I just want to use it

Install in 5 minutes, run your first task with a guided walkthrough, copy-paste from 10 everyday recipes.

User Guide → 🛠️

I want to build on it

Add a custom tool in ~30 lines, turn any OpenAI-compatible endpoint into a first-class provider, extend via MCP with zero agent code.

Developer Tutorials →

Four pillars

🧠

Agent Brain

Task tree + state machine; clean-context research sub-agents; long-term memory across sessions.

  • planning→executing→verifying
  • Three-level review
  • /mem memory
🛠️

Developer Tools

File CRUD, colored diffs, AST symbol index; snapshot before every write, /undo anytime.

  • code_index incremental index
  • Diff visualization
  • /undo
☁️

Model Ecosystem

16+ cloud providers and local models behind one interface; /model fetches lists online; free tiers work out of the box.

  • /model --fetch
  • Ollama · LM Studio
  • offline capable
🛡️

Safety System

Network sandbox, egress guard, double confirmation, full-task rollback — defense in depth, four layers.

  • unshare -n
  • egress guard
  • auto rollback

Four-layer architecture

One responsibility per layer, every layer extensible. Deep dive →

1

Agent Brain

task planning · state machine · sub-agents

↓
2

Developer Tools

file CRUD · diff/undo · code index

↓
3

Model Ecosystem

16+ cloud providers · local Ollama/LM Studio

↓
4

Safety System

sandbox · egress guard · confirm · rollback

Safety is the default, not an option

An agent that runs commands must have a constrained execution environment. Four layers ship on by default, each independently auditable.

1

Network sandbox

unshare -n

2

Egress guard

curl POST · scp · git push

3

Dangerous confirm

rm -rf · dd · fork bomb

4

Rollback

/undo · auto task rollback

Read the safety whitepaper →

By the numbers

16+providers
4safety layers
<1scold start
~50msindex update
0telemetry

Up and running in 30 seconds

git clone https://github.com/uuzjw/wow-agent && cd wow-agent && uv sync && uv run wow
Walk through your first task →