An AI QA agent that learns your app
Your users report bugs in plain English. FixFirstly signs into your app, reproduces the bug, and files a verified GitHub issue — and because it maps your app on every run, each investigation is faster and more targeted than the last.
Free during early access. No credit card. We onboard a few teams at a time.
You don't need a QA team. You need a verified bug.
Between 1 and 15 engineers, no QA hire, and an inbox full of “it's broken.” You lose half a day reproducing a report just to find out if it's even real. FixFirstly does that triage for you — it proves the bug exists before it hits your sprint. And because it learns your app as it works, every investigation after the first is faster.
I built this because I was the engineer doing that triage at 11pm on a Tuesday.
— the founder
Watch the agent reproduce a real bug.
Twelve users said checkout breaks on the second card attempt. The agent signs into your staging app, runs the steps, and comes back with a session replay and a stack trace.
12 user emails, all describing the same checkout failure, collapse into one verified bug.
The agent signs into your app with scoped test creds, walks the flow, and records every step.
“Submit clicked — no network request observed.”
2. Add item to cart
3. Go to /checkout
4. Enter card details
5. Click Pay — submit fires, no request
A GitHub issue lands in your repo with repro steps, console logs, and a replay link.
Illustration of the run — see a real recording at the top of the page.
↻ Run it again next week and it skips the exploring — it already knows where your checkout lives.
Test-case authoring is a tax engineers pay because their users won't. Your users already wrote the test cases. They're in your inbox.
The first run explores. Every run after is sharper.
As the agent investigates a bug, it builds a private map of your app — which page each feature lives on, what it's called, how to get there. Next time a report points at that area, it skips the blind exploration and goes straight to the screen. It tracks which memories actually helped, drops the ones that don't, and re-maps itself when you ship changes. Your map never leaves your workspace.
No memory yet — explores to find where checkout lives.
Jumps straight there from memory — no wasted exploration.
Every investigation leaves a memory: where each feature lives, plus a short description. Repeat runs read that memory first — and when your app changes, the map self-heals.
Every run on your app makes the agent better at your app. A tool that started today starts from zero.
The agent runs on a leash. You hold it.
Scoped test account
You give the agent one test user with the permissions you choose. It never touches a real account.
Staging by default
Point it at staging. Production access is opt-in per workspace and requires a second confirmation.
Encrypted, never logged
Test credentials are encrypted at rest. They never appear in logs, traces, or replays.
Full audit trail
Every run ships with a transcript, a session replay, and a list of every network call the agent made.
You don't trust the bot. You check its work.
The agent only files a bug it actually reproduced — and it hands you the receipts so you can confirm it yourself before anything reaches your repo.
Transcript
Every step the agent took, in order — what it clicked, typed, and observed.
Session replay
Watch the exact browser session the agent ran, frame by frame.
Network calls
The full request list — see what fired, what failed, and what never fired at all.
Or an honest miss
It retries up to 3× under different load. If the bug still doesn't show, it says so — instead of filing noise.
2. Add item to cart, go to /checkout
3. Enter card details, click Pay
4. Submit handler fires — no network request observed
Illustration of how the agent formats a verified issue — attached with the transcript, replay, and original user quotes.
Four steps. No new workflow.
Connect your inbox
Plug in Gmail, drop a CSV, or POST to the public API. Bug reports start flowing in minutes.
Cluster and prioritize
Twelve reports of the same checkout bug collapse into one cluster, scored by frequency, severity, and revenue impact.
Dispatch the agent
The AI QA agent signs into your app, walks the reported steps, and reproduces the bug on staging. On repeat runs it jumps straight to the right page — it remembers your app.
Verified GitHub issue
A clean issue lands in your repo with repro steps, a session replay, console logs, and the original user quotes — or an honest “couldn’t reproduce” if the bug doesn’t show.
Your inbox in. Verified issues out.
Bug reports come in from Gmail, CSV, or your API. FixFirstly clusters and reproduces them, then sends the verified bugs straight to GitHub — no new workflow to adopt.
The boring parts, done right.
Learns your app
Every investigation builds a private map of where your features live. Repeat runs skip the exploring and go straight to the bug — and it self-heals when your UI changes.
AI classification
Every message is tagged bug, feature request, confusion, or billing — automatically, with sentiment and severity.
Semantic clustering
Vector embeddings group the same bug across different wordings, languages, and angry tones.
Priority scoring
Clusters are ranked by frequency, severity, and the plan tier of the users complaining.
Gmail, CSV, API
Ingest support email directly, upload a Zendesk export, or POST from your existing helpdesk.
GitHub integration
Verified bugs become issues in the repo you pick, with the labels and assignees you set.
Daily brief email
One email at 9am with the top five clusters, new reproductions, and what to fix first.
Public API
Pipe messages in and pull verified bugs out. Auth via API key, documented, rate-limited.
Bring your own LLM key
SOONPlug in your OpenAI or Anthropic key and run analysis on your own quota. Shipping next quarter.
Pay for runs, not seats.
Investigations (autonomous QA agent runs) are the metered unit. Feedback ingestion is unlimited on every tier.
Free
Good enough to ship a real bug fix this week.
- 3 investigations per month
- Unlimited feedback ingestion
- Gmail, CSV, and API ingestion
- Clustering and priority scoring
- Daily brief email
- Community support
Starter
For the engineer who's also doing support.
- 25 investigations per month
- Everything in Free
- Unlimited feedback ingestion
- GitHub issue filing
- Session replays retained 30 days
- Email support, 24-hour response
Pro
For post-PMF teams shipping weekly.
- 75 investigations per month
- Everything in Starter
- Priority investigation queue
- Public API access
- Slack alerts on new verified bugs
- Priority support, 4-hour response
Scale
For teams running QA at volume.
- 250 investigations per month
- Everything in Pro
- Concurrent investigations
- Highest-priority queue
- Dedicated support
Free during early access. No credit card.
Questions engineers ask.
You create a dedicated test user with the permissions you choose. Credentials are encrypted at rest, never logged, and used only inside a scoped sandbox. The agent cannot promote itself to other accounts and every login appears in your audit trail.
Stop guessing which bug is real.
Connect Gmail, paste a staging URL, and watch the agent reproduce a bug — then watch it get faster the next time. As soon as we open access.
Free during early access. No credit card.