DevSENSEI 🧠
DevSENSEI is a strict, no-nonsense Socratic programming mentor built right into Visual Studio Code. Rather than giving you answers or writing code for you, DevSENSEI acts as a ruthlessly honest senior developer: it actively scans your code for hidden architectural flaws — memory leaks, race conditions, infinite loop risks, unhandled edge cases — and forces you to discover and reason through them yourself via targeted Socratic questions. Praise is earned, not given by default.
🚀 Features
- Inline Provider Switcher: A
Mistral | Gemini | Claude button bar sits directly in the chat panel — switch providers in one click. A green dot means the key is set; orange means it's missing.
- Zero-Friction Setup: Click a provider with no key → paste your key directly in the chat panel → ready. No settings menus to navigate.
- Multi-Provider AI Support: Full SSE streaming support for Mistral AI, Google Gemini (2.5 Flash), and Anthropic Claude.
- Real-Time Token Streaming: Watch DevSENSEI's guidance appear word-by-word instantly as it is generated.
- Zero-Tolerance Anti-Sycophancy: DevSENSEI never praises code that has flaws. It scans for hidden architectural issues — memory leaks, race conditions, unhandled edge cases — and opens with a sharp Socratic question when it finds one.
- Strict Socratic Mentoring: Guided dialogue where the AI forces you to discover bugs and reasoning errors yourself through targeted, critical questions — no empty validation.
- Custom Settings Panel: ⚙️ button in the chat header opens a dedicated settings panel — works in all IDEs.
- Premium Customizable Settings: Mentorship style (Strict / Balanced / Direct), language lock, and context history limit.
- Universal Keybindings:
Cmd+Alt+A (macOS) / Ctrl+Alt+A (Windows) on any code selection.
- Beautiful Dark-Themed Chat Panel: Clean, responsive webview that works at any sidebar width.
🔑 Setup — Get Your Free API Key
DevSENSEI requires your own API key. All three providers offer a free tier to get started:
⚙️ Extension Settings
Open the ⚙️ button in the chat header (or run DevSENSEI: Настройки from the Command Palette) to configure:
| Setting |
Description |
Default |
devsensei.aiProvider |
AI provider: mistral, gemini, or claude |
mistral |
devsensei.mistralApiKey |
Your Mistral AI API key |
(empty) |
devsensei.mistralModel |
Mistral model to use |
codestral-latest |
devsensei.geminiApiKey |
Your Google Gemini API key |
(empty) |
devsensei.geminiModel |
Gemini model to use |
gemini-2.5-flash |
devsensei.claudeApiKey |
Your Anthropic Claude API key |
(empty) |
devsensei.claudeModel |
Claude model to use |
claude-haiku-4-5-20251001 |
devsensei.mentorshipLevel |
strict / balanced / direct |
strict |
devsensei.language |
auto / bg / en |
auto |
devsensei.maxHistory |
Max messages kept in context (5–50) |
15 |
🛠️ Getting Started
- Install the extension.
- Open any file and press
Cmd+Alt+A / Ctrl+Alt+A on a code selection (or right-click → Попитай DevSENSEI).
- In the chat panel, click the provider you want (Mistral, Gemini, or Claude).
- If the dot is orange (no key yet) — paste your API key directly in the input that appears and hit Enter.
- Done — start the Socratic dialogue!
📝 Release Notes
1.3.4 — English UI & Bug Fix
- Critical bug fixed: unescaped apostrophe in the webview script was crashing the entire UI (all buttons broken). Fixed.
- Full English UI: all labels, buttons, placeholders, error messages, and settings are now in English.
- AI responds in whatever language the student writes in — no longer defaults to Bulgarian.
- Gemini default model updated to
gemini-2.5-flash.
1.3.3 — Zero-Tolerance Anti-Sycophancy
- System prompt overhauled across all three mentorship modes (Strict / Balanced / Direct).
- AI now actively scans for architectural flaws (memory leaks, race conditions, infinite loops, edge cases) on every response — not just syntax errors.
- Praise phrases ("Great job", "This is correct", etc.) are forbidden when any flaw is detected.
- When a flaw is found, response must start with a targeted Socratic question — question first, explanation never volunteered.
- Persona hardened to "strict, no-nonsense but fair senior developer".
1.3.2 — Inline Provider Switcher & Latest Models
- Provider bar in the chat panel: one-click switching between Mistral, Gemini, and Claude.
- Inline API key entry: click a provider with no key → paste directly in chat, no settings panel needed.
- Custom ⚙️ Settings panel: works in all IDEs (including Antigravity IDE).
- Gemini 2.5 Flash is now the default Gemini model.
1.3.0 — Multi-Provider Support
- Multi-Provider AI: Now supports Mistral AI, Google Gemini, and Anthropic Claude with full SSE streaming for all three.
- Bring Your Own Key: Removed the bundled fallback API key — each user provides their own key, keeping costs and security transparent.
- API Key Guard: Clear prompt with "Open Settings" button if no key is configured.
1.2.1
- Enforced informal Bulgarian tone ("ти", "твой") in all responses.
- Stateful conversation history with no chat amnesia.
1.2.0
- Premium Settings: Language lock, mentorship style, and context history limit.
- Universal Keybindings:
Ctrl+Alt+A / Cmd+Alt+A.
1.1.0
- Instant Response Streaming (SSE): Token-by-token real-time response streaming.
- Highly Responsive UI: Bulletproof CSS layout supporting sidebars down to 300px.
1.0.2
- Initial release of DevSENSEI.
| |
