Wrench.ai v4.1.0 — Stability & Performance

Stability & Performance Focus

Wrench.ai v4.1.0 is live with significant improvements to platform reliability and user experience. This release addresses critical performance bottlenecks, improves assistant sharing, and delivers a faster, more responsive interface across all workspace sizes.

TL;DR

Health checks enabled: Platform now detects and prevents request queuing and worker starvation
Dashboard performance: Eliminated 50+ redundant API calls on load — dashboards render instantly
Creative analytics: Perceptual hashing detects duplicate images; each creative now receives one consistent score
Auth self-healing: WorkOS account recovery is automatic — no more permanent 403 errors
Trial onboarding: New users receive 14 days and 1,000 credits automatically

What Changed

Backend Reliability

Health Checks: API health checks now run continuously. The system detects when a single long-running request is starving other workers and prevents the cascade that made everything feel slow and broken.
Auth Self-Healing: When a WorkOS account loses its client_id (during org migrations), authentication now automatically recovers instead of returning a permanent 403. Users regain access without intervention.
Trial Automation: New accounts receive a 14-day trial with 1,000 credits on signup — no manual activation required.
Async Processing: Creative runs and summaries now process asynchronously, freeing up request handlers immediately instead of blocking for 30-50 seconds.
Query Performance: Firmographics lookups dropped from 800ms to sub-millisecond — a 1000x improvement that compounds across all dashboard loads.

Frontend & UX

Dashboard Speed: Eliminated 50+ redundant API calls that were firing on dashboard load. Dashboards now render instantly without the previous lag.
Chat Modal: Chat interface now lazy-loads, reducing initial page load time.
Creative Polling: Fixed end-to-end polling behavior so creative scores update reliably as analysis completes.
Analytics Charts: Lead score, segmentation, and creative charts now render with correct data and accurate scaling across all views.
Error Handling: All 403 (permission denied) flows now use a consistent, branded error page with clear next steps.
Billing Optimization: Eliminated unnecessary plan check on every session load, reducing backend pressure.

Creative Intelligence

Duplicate Detection: Perceptual hashing identifies when the same image (thumbnail vs. full-size, same creative across campaigns) is submitted multiple times. Each unique image receives one consistent score instead of varying results.
Lambda Reliability: Fixed handling of recent OpenAI API changes; added comprehensive test coverage (80+ tests) for orchestration, invocation, and quota logic.

What This Means for You

For Everyone

Everything loads faster and stays responsive. No more “loading spinners” on simple operations.
Creative scores are now consistent — the same image always returns the same score.
Fewer timeout errors and transient failures from request queue backups.

For Sales Teams

New trial signups are automatic and immediate — no manual account activation.
Dashboard loads instantly, so you spend less time waiting and more time analyzing leads.
Creative analytics are accurate and reliable for pitch meetings and campaign reviews.

For Creative Teams

Ad creative scores are stable and reproducible — A/B test results are trustworthy.
Duplicate detection cleans up your library, so you see real performance data instead of inflated counts.
Async processing means you get results faster without tying up the platform.

For Administrators

WorkOS account issues resolve automatically — no more manual fixes for identity migration fallout.
Platform stability is dramatically improved. The system now prevents request starvation and worker exhaustion.
Operational overhead is reduced with automatic trial provisioning.

Technical Highlights

Why the focus on stability? The old API lacked health checks. When we added them to the new API, we immediately discovered two critical issues: (1) the dashboard was making 50+ redundant calls that could crash the server, and (2) creative endpoints were holding connections open for 30-50 seconds synchronously, starving all other workers. A lot of what looks like “frontend cleanup” was actually essential infrastructure work to keep the platform alive under real load.

Every component was stress-tested on QA before production deployment. A 503 error that occurred during testing revealed defensive gaps; all of those were hardened and re-tested before going live.

What’s Next

Cancel in-flight requests when users navigate away — reduces unnecessary backend load
Rate limiting on heavy endpoints — protects the platform from accidental overload and intentional abuse
Release notes and public documentation

Deployed April 22, 2026. Stress-tested and running smoothly on production.