DevWit: an auditable AI-native IDE for controlled code automation
DevWit runs multi-agent AI plans that propose code changes, then pauses for explicit approval before writing files or running shell commands. Created by eeyzs1 as a standalone desktop IDE, the app exposes full LLM context, terminal output, and per-item token tracking for audit. It supports TypeScript and Python language servers, integrated Git and debugging, and targets engineers and researchers who require transparent, local-first AI-assisted development workflows.
What does DevWit do and how does it operate?
DevWit accepts a developer's intent, generates a multi-step plan with coordinated agents, and requires user authorization before executing any file writes or terminal commands. The app shows system prompts, tool usage, and terminal output in a visible context manifest that can be exported as JSON, which gives developers a traceable record of every AI interaction.
How does it affect system performance during editing and parsing?
The tool uses a custom piece-table buffer plus Canvas-based rendering and tree-sitter for code parsing, which the developer notes improves rendering and edit responsiveness. Those choices indicate the app prioritizes low-latency text operations and high-performance rendering during large files and rapid edits, reducing UI lag compared with text-area based renderers.
Is it safe to let AI operate on my codebase?
DevWit enforces a granular authorization gate that stops AI agents from performing file writes or shell commands until the user approves each action. The app also exposes full LLM request context and offers per-item token tracking, enabling compliance workflows and manual review before any change reaches the repository.
Do I need technical knowledge to use DevWit effectively?
The app targets software engineers and AI researchers and ships built-in TypeScript and Python LSP support, integrated Git, and DAP debugging, which serve experienced developers. The multi-agent orchestration and explicit authorization steps require some familiarity with development workflows; casual users should expect a learning curve to interpret manifests and approve coordinated agent plans.
Recommended for developers who need auditable, local AI actions
The tool suits engineers and researchers who want direct visibility into AI decisions and strict control over execution. Expect a steeper learning curve for non-technical users, but the app's local-first design and explicit authorization model give precise operational control. Recommended.
