Describe a bug in plain English. I'll go straight to the live app and try to reproduce it — streaming each step as I go (navigating, locating the control, clicking, watching the result) — then show a diagnosis and generate a ticket draft. I only stop to ask if I get stuck (e.g. I can't find the control). Display only — nothing is filed. Follow-up messages are treated as replies — say "new bug" to start over.
Paste a chunk of unstructured text — an audio-recording transcript, a Slack conversation, meeting notes — and I'll scan it for anything that sounds like it needs a ticket (action items, bugs, requests, follow-ups) and summarize each as a candidate in the panel on the right. Click a candidate to open it; from there you can explicitly generate a full ticket draft from its guts. Display only — nothing is filed, and a draft is produced only when you click Generate. Your text lives only in this tab while it's open; a page refresh clears it (it is never saved).
Notes that help interpret the transcript (context, who's who, what a term means) and/or how to structure the output (e.g. group by project, one ticket per bug, priority hints, formatting guidance). Used together with the transcript above; leave blank to extract from the transcript alone.
Teach and bound the ComponentCRM agent. Whatever you save here becomes persistent standing context the agent applies whenever it triages, reproduces, or explores — across the Triage Chat and Text Dump tabs. Saved on the server and reapplied on every run (it survives refreshes). This only configures the agent; it never files a ticket — nothing is created without your explicit direction.
Background the agent should always know — what the app is, who the users are, domain terms, environment notes. Persistent knowledge it carries into every investigation.
Standing instructions on what to investigate and capture when reproducing — areas to probe, evidence to collect, what a good repro write-up should include.
Scope boundaries — what NOT to touch, what to stay away from, hard limits on the agent's behavior.
Turn a directive or exploration task (e.g. “explore + document consignment”) into a tracked ticket draft so the work item is captured in tix. Prefilled from “Things to explore & document” above — edit it to whatever you want the ticket to cover, then click Generate. Runs through the same draft path as the Triage Chat. Display only — a draft is produced only when you click, and nothing is filed.
Send the agent to actively explore a topic (e.g. consignment) on the live dev app. It navigates read-only, learns what the topic is, where it lives, and the screens/terms involved, then documents the findings and adds them to the agent knowledge base — standing knowledge it carries into every future triage & reproduction. Streams each step as it works. Read-only & non-destructive — nothing is changed, no ticket is filed.
Knowledge Base 0
Topics the agent has explored & documented. Click one to read the full write-up. Every entry here is injected as standing knowledge into the agent's triage, reproduction, and extraction prompts.
No explorations yet. Explore a topic above to build the KB.
Ask anything about ComponentCRM. The expert agent investigates by using tools in a loop — browsing the live dev app (read-only), querying the dev API (read-only), reading the codebase (when available), and consulting its knowledge base — and you'll see it think and act in real time. Read-only by default: it never changes CCRM data; any mutating action is only proposed for your confirmation, never executed. Every tool call is audit-logged and scoped to CCRM only; credentials are never exposed. Advisory only — it does not file tickets. Say "new question" to start a fresh conversation.
Call recordings and their Whisper transcripts. Click a recording to view its transcript, then use Send to Text Dump to extract tickets from it.