SolveAI turns plain-language instructions into executable tests, runs them across your software stack, and helps diagnose and repair failures as your product evolves.
Instruction
Generated steps
Execution log
Waiting for run…
Scripted product preview — not a live run.
01 · AI Test Builder
Describe a test the way you would explain it to a teammate. SolveAI turns it into ordered steps you can read and edit, tells you which credentials and environment values it needs, and asks when something is ambiguous — before a single browser opens.
Interpret
Parses intent, target pages, actions and the assertion that defines success.
Plan
Emits inspectable steps for a supported runtime — Playwright for UI, HTTP for APIs.
Resolve inputs
Lists required secrets and variables and asks for anything missing.
Review & save
You edit, approve and save to a project. Every change is versioned.
Start from a template
6 templatesLogin
Sign in with the test account and land on the dashboard.
Sign-up
Create an account with a fresh email and verify the welcome screen.
Checkout
Add two items, pay with the test card, expect an order number.
API health
GET /health on every service; all must return 200 under 500ms.
Form validation
Submit the contact form empty and expect three inline errors.
Regression pack
Re-run the five journeys tagged critical before each release.
// SolveAI asks before guessing
? Which account should sign in — TEST_USER or TEST_ADMIN?
? “Confirmation” — banner, toast, or email?
One workflow model
UI, API and command-line checks share the same workflow, run history and reporting — so a single journey can sign in through the browser, call your API and confirm a background job finished.
02 · Browser & UI
In developmentChromium sessions driven by Playwright. Each step records a screenshot, and failures keep the trace, console output and failed network calls.
await page.getByRole("button", { name: /upgrade/i }).click()
await expect(page.getByRole("status")).toBeVisible()
03 · API & integration
In developmentChain requests, assert on response shape and timing, and verify webhooks arrive with the payload you expect.
POST /v1/orders → 201 143ms
expect body.status == "paid"
expect webhook order.paid within 30s
04 · CLI & background jobs
PlannedRun allow-listed commands and check async jobs complete. Workers are isolated, time-boxed and cannot reach your host or private network.
$ acme-cli login --token $CI_TOKEN
$ acme-cli export --format csv
expect exit 0 · stdout ~ /rows: \d+/
05 · AI-assisted debugging
When a workflow fails, SolveAI reads the evidence — step logs, DOM state, network responses, console errors and previous runs — and classifies the failure before anyone opens a trace. You get a short explanation, the lines that matter and a suggested next move.
Application regression
The product behaved wrongly; the test was right.
Selector or layout change
An element moved or was renamed; behaviour is intact.
Environment / config
Missing variable, expired secret, wrong base URL.
Timing / synchronisation
The app was slow or the wait condition was wrong.
API contract change
Response shape or status changed from the expected contract.
Intermittent
Fails some runs and passes on identical retries.
Infrastructure
Worker, browser or network failure on our side.
// step 3 · Go to Settings → Billing
- await page.getByRole("link", { name: "Billing" }).click()
+ await page.getByRole("tab", { name: "Billing & plans" }).click()
await expect(page).toHaveURL(/settings\/billing/)
06 · Self-healing
Tests drift as products change. SolveAI proposes the smallest fix that restores the original intent and shows it as a diff.
07 · 08 · Always on
Run the same workflows continuously against production and as a gate on every pull request — one suite, two jobs.
Per-environment schedules with alerts when a passing workflow starts failing — not on every flake.
Trigger workflows from pull requests and deploys; results come back as status checks that can block a merge.
09 · Reports
Each run produces a structured report linked to its project and workflow: the failed step, the exact error, artifacts, reproduction steps, the diagnosis and a remediation hint. Filter, compare against the last green run, and export.
10 · Integrations
Source control, CI, alerts, issue trackers and a CLI, API and MCP server for coding agents. None are connected yet — each shows its real status.
11 · Security & reliability
A testing platform executes instructions against real systems. These guarantees are design requirements for the app, not add-ons.
Verified targets only
Workflows can run only against domains you have verified for a project.
Isolated workers
Each run gets a fresh, time-boxed sandbox with CPU and memory limits.
Egress controls
Private ranges and cloud metadata endpoints are blocked to prevent SSRF.
Encrypted secrets
Encrypted at rest, injected at run time, redacted from logs and reports.
Least-privilege access
Roles per team and project, enforced on the server for every request.
Audit trail
Repairs, secret changes, role changes and deletions are logged immutably.
12 · Pricing
Simple plans based on projects and monthly runs.
Free
For individuals trying SolveAI on a side project.
$0forever
Join early accessPro
For developers shipping production software.
$49per month
Join early accessTeam
For teams sharing suites across environments.
$199per month
Join early accessEnterprise
For organisations with security and scale needs.
Custom
Talk to usLaunch pricing preview. Billing is not live yet — no card is taken during early access, and limits may change before general availability.
13 · FAQ
SolveAI is in early access and the application is being built in public stages. The marketing site marks every capability as Available, Beta, In development or Planned so you always know what exists. Joining early access gets you in as each stage ships.
SolveAI uses Playwright under the hood for browser runs, but you describe the journey in plain language and review the generated steps instead of maintaining selectors by hand. The same workflow model also covers API calls and command-line checks, and failures come with a diagnosis rather than just a stack trace.
No. Before any repair is proposed, the failure is classified. Product regressions are reported as regressions and never auto-repaired. When a test itself is out of date, the fix is shown as a diff, requires approval, keeps the original version, and is written to the audit log.
Only targets you have verified for your project. Workers run in isolation with timeouts and resource limits, block private network ranges and cloud metadata endpoints by default, and never receive secrets in plain text inside AI prompts.
Secrets are encrypted at rest, scoped to a project environment, injected into the worker at run time, and redacted from logs, reports and screenshots metadata.
No. SolveAI is a conventional software product. There is no token, smart contract or on-chain component.
GitHub Actions and GitLab CI are first on the roadmap, followed by a generic API-driven integration that works with any CI system able to make an HTTP request.
Join early access and be first in as the builder, runner and reports go live.