Codero-IDE

Codero-IDE is an agentic coding workspace owned by codero-sus. This IDE is proprietary/closed-source; no open-source license is added here. Its sibling Codero-sus projects are open source, as confirmed by the owner.

Agentic coding foundation

This first working slice establishes the browser workspace and desktop shell:

• Dark, responsive IDE layout with file explorer, tabs, syntax-highlighted editor, command palette, status bar, and keyboard shortcuts.
• Local starter workspace persisted in the current browser. The starter HTML/CSS/JavaScript app runs in a sandboxed preview iframe.
• Small read-only browser terminal (help, pwd, ls, cat, clear). It deliberately does not execute arbitrary shell commands.
• Agent-first Ask, Plan, Agent, and Quest modes with OpenAI-compatible, Anthropic, Gemini, and local Ollama provider profiles. Provider keys stay on the server. Quest returns a bounded goal/requirements/design/acceptance/steps/risks spec alongside a proposed patch; read-only specialists can still return specs without edits.
• Proposed replacements include bounded per-file line diffs before approval. Choose which files are in the one task approval; excluded files stay out of both the initial patch and automatic repair passes. Applied tasks show file/line totals with Review and Undo; undo checks in-memory and on-disk baselines for later edits and safely removes task-created files. Quest specs and delivery steps can be saved to the existing Mind Desk as a local note and tasks.
• Task approval gates multi-file agent edits. In desktop mode, one approval can authorize the patch plus allowlisted local test/build checks; the agent can make up to two repair passes. Commands outside that allowlist are never auto-run: the desktop can request separate native confirmation for one plain command at a time (no shell operators or quoting), with a 120-second timeout and bounded output. Browser mode never executes host commands.
• Codero Mind adapts owner-authorized capabilities from CORTEX/AGI and MyGPT: specialist profiles with capability gates, opt-in local memory recall, persistent notes/tasks/goals, source-linked Wikipedia research, and chat-history import for ChatGPT/Claude/Telegram/WhatsApp/JSONL/CSV/ZIP. Import parsing uses Python's standard library; imported text is not used for training and is sent to a model only when memory recall is explicitly enabled.
• Codero Repo Atlas builds a local project map from the files already loaded in the workspace: languages, stack manifests, entry points, directory shape, scripts, project guidance, and relative module links. Full-text search finds matching source lines, and assistant context is ranked against the current request instead of taking the first files alphabetically. .env, credentials, generated output, and dependencies stay excluded; indexing does not make a network request.
• Python/PySide6 desktop shell that hosts the shared React IDE, exposes a narrow Qt WebChannel bridge for folder open/save and guarded local checks, and keeps workspace file/command permissions in native Python code.

The 400+ advanced-feature direction is captured as 440 scoped items across 22 areas in ROADMAP.md. It is a multi-phase product direction, not a feature count this initial slice claims to have completed. The current checkout had no application code, so this establishes the foundation to expand from.

Run in a browser

Requires Node.js 20 or newer.

npm install
npm run dev

The server binds to 0.0.0.0:4173 by default. Open the address printed in the terminal. Use Run preview (or Ctrl/Cmd + Enter) to render the current index.html, src/style.css, and src/main.js in an isolated iframe. Browser workspaces stay in browser storage; they are not written to the host filesystem.

Connect an AI provider

Copy .env.example to .env, choose AI_PROVIDER, and fill in the matching server-side key. Supported profiles are openai-compatible, anthropic, gemini, and ollama; the settings panel shows a provider-specific snippet. For example:

AI_PROVIDER=openai-compatible
OPENAI_BASE_URL=https://api.openai.com/v1
OPENAI_MODEL=gpt-4o-mini
OPENAI_API_KEY=your-key-here

Restart npm run dev after changing the environment. Provider keys are read only by the local server and never put in frontend code or browser storage. Ollama uses a local OpenAI-compatible endpoint such as http://127.0.0.1:11434/v1 and can run without a key.

Desktop

The selected desktop UI is Python 3.11+ with PySide6/Qt WebEngine. It hosts the same React IDE used in the browser, while native Python owns folder access and command approvals.

npm install
python -m pip install -r requirements-desktop.txt
npm run desktop

npm run desktop builds the browser bundle, starts the local Node API on a loopback port, and opens the Python/Qt window. For a hot-reload desktop session, run npm run desktop:dev. Use the project selector (or File → Open project folder) to choose a workspace. Text files are loaded and written through a constrained Qt WebChannel bridge; .env, credentials, binaries, symlinks, and large/dependency folders are excluded. Provider settings and Mind data live in the OS application-data directory. Out-of-allowlist commands require a native confirmation dialog.

A legacy Electron shell is retained as an optional development fallback with npm run desktop:electron. Python/Qt installer packaging is not yet configured; the desktop command currently runs from a source checkout with Node.js available.

Development checks

npm test
npm run test:python
npm run build

The Node tests cover persistent Mind data, capability filtering, Wikipedia redirect boundaries, the provider/import API (including Quest output and malformed-response fallbacks), bounded patch-diff previews and file-selection safeguards, source search/ranking, and Repo Atlas generation. Python tests cover chat-export parsing and desktop workspace/command boundaries.

Capability sources and scope

The repository owner authorized these open-source companion projects as feature references; Codero-IDE itself remains proprietary and has no open-source license added:

• codero-sus/agi — named, capability-scoped agents; research with linked sources; long-term memory, notes, and goals.
• codero-sus/MyGPT — chat-export parsing, episodic recall, feedback-oriented memory, and local Mind/Growth workflow patterns.

Codero adapts the useful product capabilities rather than importing either project's model-training runtime. Chat import uses the Python standard-library parser; memory is local and opt-in for provider requests. Wikipedia is the only live research host in this slice. Dynamic skill-code execution, self-modifying skills, and automatic weight training are deliberately not enabled.