GitHub18m ago
Deploy log
Two changes in the window.
Reassemble a failing system's history in one query.
At 2 a.m., there is no time to page whoever built it. Virgo reassembles the relevant work history (what changed, the constraint it was protecting, the decision behind it). You debug the real system rather than your best guess.
sourcesPR #5012PAY-77#paymentscfg-8841
›
⏎ send · esc to interrupt
The story is scattered across tools and months
The deploy is in GitHub, the ticket is in Jira, and the reason is in a Slack thread from May. Each piece is easy to find alone. Nothing connects them, and at 2am there is no one to ask.
Which one?
Noise
Slacknow
#incidents
Scrolling fast.
Unlinked
JiraMay 19
PAY-77
The cap ticket.
Is this pool cap load-bearing, or safe to revert?
“Start digging.”
Buried
SlackMay 16
#payments
The reason, somewhere upthread.
Unread at 2am
ConfluenceMay 22
Pool sizing decision
Forty pages of context.
From May
Google MeetMay 12
Outage retro
Recorded, unwatched.
One question returns the connected trail.
Behind one question, Virgo walks the reasoning graph from the symptom to the change to the decision it protects, and returns the picture already assembled.
›
⏎ send · esc to interrupt
Give the paged-in agent the whole trail behind the outage.
Outage history reaches the on-call engineer and the agent paged in beside them.
For engineers
Debug the real system with its history attached.
The change, the constraint it protects, and the incident that created it arrive connected. You know whether a fix is safe before you ship it, while the page is still open.
- What changed: the deploy and the PR behind it.
- What it protects: the constraint that made the change load-bearing.
- What happened before: the prior incident on the record.
For AI agents
The agent respects what the system is protecting.
An agent handed an incident with no history suggests the rollback that reintroduces last month’s outage. With the assembled picture, it reasons from the real constraint. Virgo retrieves that context across the graph at query time, over MCP.
97%context retrieval accuracy
Early evaluations reached 97% context retrieval accuracy, compared with about 68% for baseline RAG or hybrid search when reassembling the history behind a failing system mid-incident.
Built on Context retrieval → and Decision lineage →