AutoHuolto AI

What it does

AutoHuolto AI is a privacy-first, stateless vehicle service-history analyzer for people reviewing used-car maintenance records. It turns user-redacted service-book pages, receipts, and maintenance photographs into an editable timeline, helps confirm the exact vehicle variant, researches maintenance intervals with visible sources, calculates maintenance status deterministically, and exports a local JSON or Excel report.

Privacy-first workflow

Images remain in browser memory while the user redacts names, addresses, registration numbers, VINs, and other identifiers. The browser generates new sanitized PNG blobs, and only user-approved sanitized images are sent for extraction. The app has no user accounts, database, object storage, persistent cache, or background jobs. The UI clearly discloses that provider retention policies may still apply to transmitted sanitized content.

How it works

1. Enter vehicle and current-odometer details.
2. Upload and redact service records locally in the browser.
3. Extract schema-validated, editable maintenance events.
4. Resolve and explicitly confirm a vehicle variant with visible sources.
5. Research source-backed maintenance intervals and preserve conflicts and uncertainty.
6. Calculate statuses in TypeScript from reviewed history and validated evidence, rather than asking the model to decide whether work is due.
7. Inspect the source-backed report and download JSON or Excel locally.

When evidence is missing or incompatible, the application reports insufficient evidence rather than inventing an interval or recommendation.

Built with Codex and GPT-5.6

Codex accelerated implementation of the Next.js application, browser-side canvas redaction, Zod schemas, API boundaries, status calculations, report exports, privacy checks, unit and component tests, and Playwright end-to-end coverage. The developer guided scope, tested workflows, reviewed behavior, and verified changes.

GPT-5.6 was used through Codex during specification and development. At runtime, a configurable OpenAI Responses API model supports bounded evidence extraction, vehicle-variant research, and maintenance-interval research. Every model output is schema-validated; OpenAI requests use store: false; and application code calculates due and overdue outcomes deterministically.

Run it

The repository README documents prerequisites, environment configuration, synthetic sample data, a no-network demo flow, and lint, privacy-audit, type-check, unit-test, end-to-end-test, and production-build commands.

Built With

• codex
• exceljs
• gpt-5.6
• next.js
• openai-responses-api
• playwright
• react
• typescript
• vitest
• zod