Track every improvement, feature, and fix
April 2026
Upload any existing project to the VFS and let the Audit Loop automatically detect and fix all errors. Supports React, Next.js, Vue, Svelte, Node.js, and static HTML. Type fix my project in chat or enable the Audit Loop toggle to get started.
Before Brain Mode starts generating, a compact card asks 4 key questions (auth, data source, mobile-first, project size). Answers are merged into the architect's spec for more accurate generation from the first run.
Memory now filters by stack type. React patterns stay separate from Vue patterns. Error taxonomy with 9 semantic categories (import_error, prop_mismatch, type_error, etc.) enables better deduplication and faster learning.
Fixed unescaped single quotes inside the Next.js stack context string that caused a browser SyntaxError ("Unexpected token 'export'") when the file was loaded as a non-module script.
March 2026
Learning context is now injected into every task execution, not just planning. Agents see successful patterns and known issues while writing code.
Custom Docker image with Python 3.11, Node.js 20 and PHP 8.2 for in-browser execution across 37+ languages without WebContainer.
Before writing a component, agents now automatically see where it is used and what data fields are passed to it — reducing prop mismatches and field inconsistencies.
Immediately after each file is saved, the system verifies data field consistency and enum map completeness across related files.
Migrated to a dedicated VPS for improved performance, reliability and response times.
Expanded template library with categories and AI-based auto-detection. New templates include Next.js SaaS, fullstack React+Node, and AI chatbot.
February 2026
Real-time file validation during generation. Auto-fixes missing React imports, export statements, and import path issues on the fly.
Validates all critical files after build — checks content validity, dependency completeness, and missing imports before starting the dev server.
Scans all generated code for import statements, detects missing npm packages, and automatically adds them to package.json.
Each agent now handles exactly one file per task, eliminating conflicts during parallel execution and improving output consistency.
Prevents multiple agents from writing to the same file simultaneously during parallel execution.
January 2026
Upload PDFs, text files or paste content to give the AI context about your project, custom APIs or frameworks. Injected automatically into every relevant task.
Free tier with daily limits, BYOK ($7.99/mo) with your own OpenRouter key, and Pro with ForgeLab token-based usage. Full payment and billing flow implemented.
Brain now retains successful patterns and known issues across sessions, improving output quality over time for each user.
Pre-built templates for React+Vite, static HTML, Next.js and more. Auto-generates missing critical files on startup.
Early January 2026
Terminal errors are automatically detected, analyzed and sent to the Bug Hunter agent. One click fixes runtime failures, broken imports and syntax errors.
Input validation, rate limiting and improved authentication flow across all API endpoints.
Each agent now has a configurable persona with a dedicated model, temperature and instructions. Custom workflows can be defined per project type.
Late December 2025
Full Node.js environment running inside the browser via StackBlitz WebContainer. Projects build and run without leaving the tab.
Dedicated orchestration agent that analyzes the user request, detects the tech stack, plans the full task graph and delegates work to specialized agents in the correct order.
Deploy any generated project directly to Cloudflare Pages from within the editor with a single click.
Full shell access inside the browser. Run npm commands, view build output and interact with the WebContainer environment directly.
November - December 2025
First working version of ForgeLab. Single-agent chat mode, Monaco code editor, virtual file system, and basic file management.
Initial multi-agent system with parallel task execution. Architect and Senior Dev agents working simultaneously on the same project.
In-memory file system with full tree navigation, create, rename, delete and export to ZIP.
All AI requests routed through OpenRouter from day one, enabling access to Claude, GPT, Gemini, DeepSeek and 200+ other models from a single API.