Guide
Why wow-agent
The old way is broken
- Context loss — copy/paste between editor and chat
- No execution — AI cannot run tests or verify changes
- You are the bridge — manual integration of every snippet
- No guardrails — suggestions applied directly, no undo
The wow-agent way
| Principle | Implementation |
|---|---|
| Agent, not assistant | Executes real operations in your repo |
| Local-first | Zero telemetry; works fully offline with local models |
| Model agnostic | 16+ cloud providers + Ollama/LM Studio |
| Safety by default | Sandbox, egress guard, confirmations, rollback |
| Extensible | MCP protocol + custom tools + sub-agents |
Comparison
| Difference | Traditional tools | wow-agent |
|---|---|---|
| Models | locked to one vendor | any OpenAI-compatible API |
| Runtime | IDE / cloud bound | pure terminal |
| Arch | x86 only | native ARM64 |
| Safety | direct execution | sandbox + rollback |
One sentence: model choice, environment control, safety decisions.