01Your agent writes code
Your agent writes code
and tests
Ask for a feature. Your agent ships the code and tests, runs them on a full copy of your app, and grows the suite with every change.
02Review scenarios
Review scenarios
instead of code
Every run is recorded. Open a new or changed test and watch what happened in the browser. No code review required.
project member roles
›
viewer can read tasks
new
WEB
Open Website launch as a viewer
0.0s
circular.test/website-launch
00:00 / 00:05
03
Build and merge with confidence
Deterministic end-to-end regression on every run. Your agent iterates to green, and you merge large changes with minimal review.
Add viewer roles
#142
Open
agent wants to merge 1 commit into
main
from
viewer-roles
agent
Add viewer roles
8f21c
Checks are running
Testing viewer permissions
Merge pull request
Test anything and
everything
Each run gets an ephemeral environment with your full stack, including code, infra, and dependencies. Everything can be tested end to end.
website launch board
›
create a task from the board
passed
3.0s
WEB
Open the Website launch board
0.84s
circular.test/website-launch
00:00 / 00:05
circular mobile
›
mark a task done from the app
passed
3.4s
MOBILE
Launch the Circular app
0.92s
00:00 / 00:05
tasks api
›
create a task and enforce permissions
passed
1.4s
HTTP
Sign in as Alex, an editor
0.21s
POST http://api:3000/auth/sessionstatus 200 · 38 ms
Request
{
"email": "[email protected]",
"password": "••••••••"
}Response
{
"user": "alex",
"role": "editor",
"token": "sess_9f2…"
}circular cli
›
create a task from the terminal
passed
3.0s
EXEC
Sign in with the CLI
0.84s
cliexit 0 · 0.8 s
$ circular login --token "$CIRCULAR_TOKEN" Logged in as [email protected] (editor)
circular mcp server
›
create a task with the agent’s tools
passed
2.1s
MCP
List the server's tools
0.18s
list tools6 tools · 9 ms
| name | description |
|---|---|
| list_tasks | Tasks on a board |
| get_task | One task by id |
| create_task | Add a task to a board |
| move_task | Change a task's column |
launch assistant
›
agent adds a task from a request
passed
2.7s
FAKE
Load the recorded model responses
0.06s
openai.replayrecorded · 2 ms
Result
{
"cassette": "recordings/announce-launch.json",
"host": "api.openai.com",
"interactions": 2
}billing
›
a paid upgrade notifies the team
passed
1.6s
HTTP
Deliver Stripe's checkout webhook
0.22s
POST http://web:3000/webhooks/stripestatus 200 · 46 ms
Request
{
"type": "checkout.session.completed",
"data": {
"customer": "cus_9Lk2",
"subscription": "sub_7Hq1"
}
}Response
{
"received": true
}Test what your users see
Browser tests click, type and navigate the way your users would, and you get a full visual replay of what they see.
Test the experience in your app
Exercise mobile user flows alongside the backend that powers them. Follow a scenario from a tap in the app to the data it creates, with your services running together in the same test environment.
Real requests. Your whole backend.
Send requests to your running services and assert on responses, database records, and side effects. Test authentication, permissions, and the paths that cross service boundaries.
Test the command line too
Run your CLI inside the test environment, the way a user or an agent would. Assert on its output and exit code, then on what it actually changed in your services and your database.
Put your tools to the test
Call your MCP tools as an agent would. Verify their responses and what actually changes in your app, with the server and its dependencies running in an isolated environment.
Test the agent, tools and all
Replay recorded model responses deterministically while running the real tool calls against your test environment. Check the result of a complete agent workflow, from its first action to the changes it makes.
Test your integrations too
Stand in for Stripe, Slack, PostHog, and any other third party with fakes that run inside the test environment. Deliver their webhooks, check what your app sent them, and test the complete flow without touching a real account.
Frequently Asked Questions
How is it different from QA agent platforms?
Your coding agent writes tests that become part of your codebase. Specific Test runs those tests deterministically against a full copy of your app, and records what happened. You can review the scenarios and replay the results every time the code changes.
Won’t the tests be slow?
Tests run in parallel on our servers, and scenarios can build on shared setup instead of starting from scratch every time. The time a suite takes depends on your stack and what you test; we help you tune it during onboarding.
How is it different from a regular test suite?
Specific Test brings your full environment into the test: your frontend, backend, databases, and other services. Tests interact with the same surfaces your users do, with recorded runs and a diff of new and changed scenarios to help you review a change.
Can I integrate it into my CI?
Yes. Run the suite from CI and use the result as a pull request check. Your agent can inspect failures, fix the code, and run the tests again before you merge.
Do I need a staging or QA environment?
Specific Test creates an isolated, ephemeral environment for each run, with a full copy of your stack. The suite does not depend on a shared staging environment or access to production data. External APIs can be represented by fakes or recorded interactions.
How much does it cost?
Specific Test is in private preview. Get in touch to discuss access, your stack, and pricing for your team.