🧪 FusionTest

AI-powered UI testing for web, desktop and mobile — built by FusionLeap.io

FusionTest runs UI tests written as plain-English goals. An LLM reads the screen and decides what to do, a verifier confirms each goal against what's actually on screen, and goals that pass are recorded so later runs replay them without calling the AI at all.

✨ Features

Feature	Notes
Natural-language goals	`"Open the Pricing page and verify all three plans are shown"`
Goal discovery	Crawls a site and proposes goals grounded in its real content
Record & replay	Passing goals replay with no LLM calls; the AI repairs them when the page changes
Honest results	Capacity or infrastructure problems are reported as errors, not test failures
Multi-provider LLMs	Groq, Claude, Gemini, OpenAI or local Ollama, with automatic fallback
Safe secrets	`{{PASSWORD}}` placeholders — the AI never sees values, which are only typed on allowed sites
Guardrails	Invalid-action checks, loop detection, backtracking, no-progress detection
Reporting	JSON reports, dashboard uploads, failure screenshots, Slack, GitHub Actions annotations, webhooks
Platforms	Web (Playwright), Electron, Android & iOS (Appium), Windows, macOS

🏗 Architecture

fusiontest/
├── core/         # Runner, LLM action model + fallback chain, goal verifier,
│                 # screen parser, secrets, replay, SSRF guard (netguard)
├── desktop/      # Playwright (web/Electron), Windows, macOS adapters
├── mobile/       # Android (UIAutomator2) and iOS (XCUITest) adapters via Appium
├── discovery/    # Goal discovery from a site's pages
├── guardrails/   # Action validation, loop detection, backtracking
├── reporting/    # JSON reports, Slack, GitHub annotations
└── training/     # Experimental MPNet fine-tuning (not wired into runs)
api/              # SaaS backend (FastAPI + PostgreSQL) — see docs/api.md
dashboard/        # React dashboard
website/          # Marketing site (Next.js)

Design decisions are recorded in docs/adr — start with ADR-007.

🚀 Quickstart

# macOS (Homebrew Python) or anywhere you want an isolated CLI — recommended
brew install pipx && pipx ensurepath        # then open a new terminal
pipx install "fusiontest[desktop]"
"$(pipx environment --value PIPX_LOCAL_VENVS)/fusiontest/bin/playwright" install chromium

# …or inside a virtual environment
python3 -m venv .venv && source .venv/bin/activate
pip install "fusiontest[desktop]"
playwright install chromium

export GROQ_API_KEY=gsk_...          # or ANTHROPIC_API_KEY / GEMINI_API_KEY / OPENAI_API_KEY

fusiontest run --url https://example.com \
  --goal "Verify the page heading says Example Domain" --headless

# No goals yet? Map what works on a test/staging site, and save the flows that do
fusiontest scan https://staging.example.com --save fusiontest.goals.yaml
# Behind a login: EMAIL and PASSWORD come from the environment, never the AI
fusiontest scan https://staging.example.com --login "Sign in with email {{EMAIL}} and password {{PASSWORD}}"

Groq, OpenAI and Ollama work out of the box; for Claude or Gemini add the extra — fusiontest[desktop,claude] or fusiontest[desktop,gemini]. Upgrade with pipx upgrade fusiontest (or pip install -U fusiontest in the venv).

Set FUSIONTEST_API_KEY (from the dashboard's API Keys page) to upload each run to your FusionTest dashboard.

Exit codes: 0 all goals passed · 1 a goal failed · 2 some goals couldn't be evaluated (LLM capacity or infrastructure) — not a test failure.

Full guide: docs/getting-started.md — platforms, goal files, configuration, secrets, replay, dashboard and CI.

📝 Writing goals

# tests/goals/smoke.yaml
name: Marketing site smoke test
platform: playwright
url: https://example.com
goals:
  - "Open the Pricing page and verify three plan tiers are shown"
  - "Open the Contact page, submit the form empty, and verify required-field errors appear"
  - name: Log in
    goal: "Log in with {{EMAIL}} and {{PASSWORD}} and verify the dashboard loads"

fusiontest run --file tests/goals/smoke.yaml --headless

Tips: say what to verify, not just where to go; one outcome per goal; don't submit real forms on production sites.

🛠 Configuration

fusiontest init creates fusiontest.config.yaml:

backend: groq               # unset = auto-detect from API keys
guardrails:
  loop_detection_window: 5
  max_retries: 3
replay:
  enabled: true
  trace_dir: .fusiontest/traces
  reverify_every: 10

🔌 CI/CD

Run fusiontest run --headless in any CI system. With FUSIONTEST_API_KEY set, replay recordings are shared through your FusionTest account (laptop, CI and dashboard all replay each other's passing goals); without it, cache .fusiontest/traces. Guides for GitHub Actions, Bitrise, Fastlane and Jenkins: docs/ci-integrations.md.

🤝 Contributing

See CONTRIBUTING.md and ROADMAP.md.

📄 License

MIT © FusionLeap.io

Metadata

Release files for fusiontest 0.2.39

For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.

Source distribution (sdist)

Source distribution for fusiontest 0.2.39
File	Size	Uploaded	Provenance
fusiontest-0.2.39.tar.gz	142.2 kB	Oct 13, 2026	Without provenance	Details

Built distribution (wheel)

Table of built distributions (wheels) for fusiontest 0.2.39
File	Interpreter	ABI	Platform	Reset
fusiontest-0.2.39-py3-none-any.whl 158.9 kB Oct 13, 2026 Without provenance	Python 3	none	any	Details

Total release size: 301.1 kB

Release files / fusiontest-0.2.39.tar.gz

Download URL	fusiontest-0.2.39.tar.gz
Size	142.2 kB
Tags	Source
SHA-256 checksum How to use checksums	`49b088e2223f1dcdaa67b766be39d2fd11ae39e72bde6bc5bc1384c8179442f8`
BLAKE2b-256 checksum How to use checksums	`3056ea4641a191ad5aec46756bd5320cf3fb7c94fc77c28df2bb989ef11d001f`
Upload date	Oct 13, 2026
Uploaded using Trusted Publishing? What is trusted publishing?	No
Uploaded via	`twine/7.0.0 CPython/3.13.14`

Release files / fusiontest-0.2.39-py3-none-any.whl

Download URL	fusiontest-0.2.39-py3-none-any.whl
Size	158.9 kB
Tags	Python 3
SHA-256 checksum How to use checksums	`6feddbe62c5490ca7543ac27eb11242ae18aba831d0d767b71091e0ab3a877d6`
BLAKE2b-256 checksum How to use checksums	`fb5537179ad38000759288235c6e3e4eb252b4622416bf04e628bc9c5e3dcc86`
Upload date	Oct 13, 2026
Uploaded using Trusted Publishing? What is trusted publishing?	No
Uploaded via	`twine/7.0.0 CPython/3.13.14`

Security for fusiontest 0.2.39

Report project as malware

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

No provenance found

None of the files in this release were uploaded with provenance metadata.

A note on package security

PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.

To learn more about protecting yourself from malicious actors, visit our package security documentation.