Lastest
AI-powered visual regression testing platform that records tests once, replays them forever in CI, and uses three diff engines to catch UI regressions before they reach production.
At a Glance
Hosted by Lastest Cloud. Kick the tires on a real project. No credit card.
Engagement
Available On
Alternatives
Listed Jul 2026
About Lastest
Lastest is a visual regression testing platform built for solo founders and small teams who ship frequently. It combines an agentic AI that generates Playwright tests automatically with a human-in-the-loop oversight dashboard, so teams get AI speed without sacrificing control. The tool is available as a self-hosted Docker deployment (FSL-1.1-ALv2 license) or as a managed cloud service.
What It Is
Lastest sits in the visual testing category alongside tools like Percy, Applitools, and Chromatic, but takes a different economic approach: AI tokens are spent once during test generation, and every subsequent CI replay runs as pure deterministic Playwright execution at zero token cost. The core workflow is: point the agent at a staging URL, let it map flows and generate tests, review and approve baselines in the dashboard, then replay on every commit. Three diff engines — pixelmatch (pixel-perfect), SSIM (structural), and Butteraugli (perceptual/human-aligned) — analyze screenshots and surface only meaningful regressions.
How the Generate-Once, Replay-Forever Model Works
The platform's economic argument centers on separating AI usage from test execution:
- Generation phase: The Play Agent (an 11-step autonomous pipeline with specialized sub-agents — Orchestrator, Planner, Scout, Diver, Generator, Healer) explores the app, maps routes, and writes resilient Playwright code with a 7-layer selector fallback (data-testid → id → role → aria-label → text → CSS → OCR).
- Review phase: Every AI-generated test lands in a pending state in the oversight dashboard. Humans approve, reject, edit, or request regeneration before tests enter the active suite.
- Replay phase: Subsequent runs are pure Playwright execution — no LLM calls, no token costs. A SHA256 fast-path skips pixel comparison entirely when screenshots are identical.
The platform also supports manual point-and-click recording (no AI or API keys required) and AI-assisted mode where humans review every AI change before it is saved.
Flaky Test Prevention and Diff Precision
Lastest ships 12 stabilization features to reduce false positives: timestamp freezing, random value seeding, cross-OS font consistency, burst capture (multi-frame stability detection), auto-masking of dynamic content, network idle waiting, DOM stability detection, third-party domain blocking, font loading wait, loading indicator hiding, animation freezing, and page-shift detection. Text-region-aware OCR diffing applies separate thresholds to text vs. non-text regions, and configurable ignore regions mask dynamic areas like timestamps or ads before comparison.
CI/CD Integration and Infrastructure
Lastest integrates natively with GitHub Actions (reusable composite action las-team/lastest/action@main), GitLab CI (including self-hosted GitLab with MR comments), and any pipeline via the @lastest/runner CLI. Smart Run analyzes git diffs and runs only the tests affected by a given change. The embedded browser runs containerized Chromium with live CDP streaming — no local Playwright install required. Remote runners connect via WebSocket for distributed execution across OS and browser targets. Scheduled cron runs, webhook triggers, and Slack/Discord notifications round out the automation surface.
MCP Server and AI Agent Integration
The platform ships an MCP (Model Context Protocol) server (npx @lastest/mcp-server) exposing 52 tools that let external AI agents — Claude, Cursor, or any MCP-compatible client — run tests, review diffs, approve baselines, create tests, and heal failures programmatically. Six AI providers are supported for test generation and diff analysis: Claude (Agent SDK, CLI, OpenRouter, direct Anthropic API), OpenAI, and Ollama (local models). A separate provider can be configured for diff analysis vs. test generation.
Current Status and Deployment Options
The GitHub repository (las-team/lastest) was created in April 2026 and last pushed in June 2026, indicating active development. The project is licensed FSL-1.1-ALv2 and the full source is on GitHub. Self-hosting requires Docker 20+, Node.js 18+, and a k3d cluster for the embedded browser stack. The cloud-hosted option (Lastest Cloud) is available for teams that want to skip infrastructure management. The platform is written in TypeScript and Rust, built on Next.js 16, React 19, PostgreSQL with Drizzle ORM, and uses axe-core for automated WCAG 2.2 AA accessibility scoring on every screenshot capture.
Community Discussions
Be the first to start a conversation about Lastest
Share your experience with Lastest, ask questions, or help others learn from your insights.
Pricing
Free
Hosted by Lastest Cloud. Kick the tires on a real project. No credit card.
- 1 project
- Shared runner pool
- 500 capped runner-minutes
- 1 concurrent run
- Community support
Self-Host
Run Lastest on your own infrastructure. Free forever, unlimited screenshots, full source on GitHub.
- Unlimited screenshots
- Full source code access
- FSL-1.1-ALv2 license
- Docker deployment
- All platform features included
Starter
For solo devs and small teams shipping regularly. Early adopter locked-in price.
- 3 projects
- 5,000 priority run-minutes
- 2 concurrent runs
- Email support
Growth
Multiple apps, CI pipelines, Slack-speed support.
- 10 projects
- 30,000 priority run-minutes
- 5 concurrent runs
- Slack support
- CI integrations (GitHub, GitLab, Bitbucket)
Pro
Capacity, SSO, and priority response for serious teams.
- Unlimited projects (reasonable quotas)
- 120,000 priority run-minutes
- 15 concurrent runs
- SSO / SAML
- Priority support
Enterprise
SLA-backed uptime, SSO/SAML, managed cloud, dedicated support, on-premise installation, and priority roadmap.
- SLA-backed 99.9% uptime
- SSO / SAML integration
- Managed cloud deployment
- Dedicated support channel
- On-premise installation support
- Priority feature development
- Custom implementation per quote
Capabilities
Key Features
- AI-powered visual test generation via Play Agent
- Three diff engines: pixelmatch, SSIM, Butteraugli
- Human-in-the-loop oversight dashboard
- Zero-token deterministic replays after initial generation
- Self-hosted Docker deployment
- GitHub Actions and GitLab CI integration
- Smart Run (git-diff based test selection)
- MCP server with 52 tools for AI agent integration
- 12 flaky-test stabilization features
- WCAG 2.2 AA accessibility scoring (axe-core)
- Autonomous 11-step Play Agent pipeline
- Manual point-and-click browser recording
- AI test fixing and auto-healing
- Spec-driven test generation from OpenAPI/user stories
- Embedded browser with live CDP streaming
- Remote distributed runners via WebSocket
- Branch baseline management (fork/merge/promote)
- Scheduled cron-based test runs
- AI failure triage and classification
- Role-based access control (Owner/Admin/Member/Viewer)
- Multi-tenant team workspaces
- Gamification: Beat the Bot leaderboard
- In-app bug reports with auto-captured context
- 6 AI providers including Ollama for local inference
- Test versioning and composition
- Selector stats and success rate tracking
- Google Sheets test data integration
- Slack, Discord, and custom webhook notifications
