Roadmap

Fires when the same error hash appears 2–5× in a session depending on sensitivity setting.

Detects when you are editing the same region of code repeatedly with no forward progress.

Free tier. Extracts imports and a focused window around the current error.

Pro tier. Full AST parse, Shannon entropy scoring, Myers delta, memory cache.

Setup path for Claude Code, Cursor, Windsurf, Codex CLI, Zed, and VS Code / Copilot.

bash, zsh, fish on Windows (WSL), macOS, Linux. 60–90% reduction on npm, git, docker output.

URI callback scheme auto-detected from vscode.env.uriScheme.

x64 and ARM64. Binary distributed inside platform-specific VSIX.

Metrics sync every 5 minutes and on session end. The web dashboard refreshes periodically for signed-in users.

Reduce the gap between local session changes and what appears in the signed-in web dashboard.

Add stronger retry and offline queue behavior so metrics are not dropped when the API is temporarily unreachable.

Keep tightening metadata, packaging, and onboarding details for smoother Marketplace review and install success.

Turn the existing paid-tier story into explicit product behavior instead of copy-only gating.

When a loop is detected, LoopGuard asks Claude to suggest three fundamentally different approaches — not just renamed versions of the same fix.

LoopGuard stores a hash of every AI prompt you send. Detects when you re-send the same prompt and warns you before you waste another turn.

Shows which files in your project are causing the most loops across all sessions. Identifies structural bugs, not just session-level issues.

Teams can route loop alerts to a shared channel. Useful for pair debugging and onboarding tracking.

Separate plugin using the IntelliJ Platform SDK. Same loop detection and context engine, different activation model.

Org-level view of loop rates, token spend, and time wasted across all developers. Identify which parts of the codebase create the most AI friction.

Set a daily token budget per developer. LoopGuard warns when 80% is consumed and blocks context copies when the limit is hit.

Route cheap queries (boilerplate, formatting) to smaller models and only send complex loop-breaking queries to GPT-4 / Claude Opus.

After each AI response, score how relevant it was. Feed this back into the entropy model to improve future context selection.

Route context-heavy requests to a local model (Ollama, LM Studio) to eliminate API costs entirely for routine tasks.

Aggregate loop patterns across all users (opt-in, anonymized) to identify the top 100 error patterns that cost developers the most time industry-wide.

Run LoopGuard in CI to detect loops in automated test runs — useful for flaky test debugging and build pipeline optimization.