GIGACHAD-QA

Run configuration

1 Target

Target URL

Or enter branch manually

Testing production — agents will avoid destructive actions

Monorepo path optional

Project mode connects your repo so findings link to exact files and lines.

2 Engine

Provider

Model

Agents?

10 parallel agents

3 Scope

Custom instructions optional

Injected verbatim into every agent's system prompt.

Context attachments prompts & reference

Reference material shared with every agent — paste or attach prompts, instructions, examples, specs, or screenshots to guide the run. Up to 3 files (txt/md ≤48KB, png/jpg ≤1.5MB); or paste an image anywhere on this card. For PDF/DOCX, paste the text instead. Note: attachments don't survive a page reload.

Agent assets one per agent

Upload up to 300 files (e.g. one resume per agent). Each agent is assigned one round-robin — upload as many as your agent count for a unique file per agent. PDF/DOC/DOCX/TXT ≤8MB each.

Logged-in testing options

Give agents test accounts to log in with (assigned one each, round-robin), or have them sign up themselves. Tell agents what to do in the instructions above. File uploads are configured above in “Agent assets”.

This is a multi-step flow / wizard (sequential screens on one URL) Logins

Passwords are sent only to the agents and scrubbed from logs, results, and recordings.

4 Limits

Budget cap USD?

Max steps per agent? Auto

Auto picks the most steps each agent can use — ~250 at low agent counts, scaling down as you add agents. Total budget ~70,000 steps/run.

Per-agent timeout seconds

Viewports

Record video per agent Cluster & deduplicate findings

More advanced

Funnels (one per line, label=url or url)

Behaviors (one per line)

QA marker param

Record Playwright traces

Recording adds storage and a longer finalize step on large swarms. Sampled uploads only videos from agents with findings, plus a random sample of up to 10 others.

The consolidator runs automatically after every swarm and writes BUG_FIXES.md.

Findings

Ready when you are.

Configure the run on the left, then dispatch the swarm. Findings stream in here the moment agents start crawling.

0 findings 10 agents queued

QA swarms on demand.

Run configuration

Findings

Ready when you are.