Engineers working on complex codebases (50k+ lines) who use AI coding tools like Cursor or Claude Code. Especially valuable if you:

Security. Your code already lives on your machine - there's no reason to clone it to a cloud sandbox. Running locally means your codebase never leaves your control, and your legal/security team can sleep soundly. No data residency concerns, no vendor lock-in, no wondering where your code is stored.

You've already vetted your AI stack with your organization. Using your own key means:

About $0.07 per ticket with GPT-4o-mini (surprisingly capable for this) or $0.30 with GPT-4o. Since better context means your implementation agents search less and derail less, you often save more than that on tokens spent on the actual implementation.

I built Superfocus for my own workflow and have never had a job that didn't use either Linear or Jira. I find issue trackers distracting and overwhelming; after every ticket I find myself having to get my head around what to work on next. Are there new tickets? What is the best critical path? What urgent tickets are waiting for me? How do new tickets fit in? Superfocus exports research output to the description of a ticket, but also exports labels to estimate impact, effort, unblocks and risk. These labels power the sorting algorithm the vscode and cursor extensions use to prioritize tickets. As a result, I safe time, feel less overwhelmed and rarely open my issue tracker.

About 5 minutes depending on codebase size and ticket complexity. But it runs in the background while you work on other tickets, so by the time you're ready for something new, the research is done.

If you just created a ticket and want to start immediately, you can skip the Superfocus prompt - but waiting saves you time in revision rounds later. Go grab lunch, answer Slack messages, review a PR. The wait is worth it.

You'll need a Personal Access Token (PAT) for whichever tracker you use:

Paste it into Superfocus settings and you're done.

You can rerun prep runs if needed. But honestly, I've tuned the Superfocus agents extensively to avoid this — I almost never need to rerun them myself. The filtering and sub-agent architecture is designed specifically to prevent the context derailment that plagues other tools.

Superfocus doesn't plan - it does unopinionated research. Think of it like a super-experienced founding engineer giving you gotchas to kickstart your work.

A possible workflow: Run Superfocus prep → use that context in Plan Mode to create an approach → implement. This actually improves Plan Mode because it starts with all the puzzle pieces already assembled instead of searching as it goes.

Superfocus makes any agent or LLM coding tool smarter by front-loading the codebase knowledge they need.

Yes. Like Cursor, ChatGPT, and other systems, you can include custom rules to guide what patterns, gotchas, or conventions Superfocus should prioritize in your codebase.

Gemini is supported as well, and more models coming soon. Let me know what you need.

Yes — but Superfocus does require VS Code or Cursor running in the background to access your codebase. You can let it sit in the background while you work in Claude Code or anywhere else.

The workflow is hands-off: Superfocus researches in the background, exports to your ticket, then you copy the prompt with one button and paste into Claude Code (or any other tool).

Yes. Superfocus only accesses what's already on your machine. Nothing gets cloned or uploaded anywhere.

Only anonymized usage metrics to improve the product, and billing information if you're on a paid plan. I never see your code, ticket content, or codebase. It all stays local or goes directly to your Linear/Jira.