The batch uses a hard-coded days=30 assumption. Finance added a workaround in 2019; the ticket explains the regulatory reason.
Memory layer for legacy systems
Code tells you what. Backstory remembers why.
Cited answers from decades of commits, tickets, and departing experts — with receipts you can click, not vibes you have to trust.
— Try the demo — pick a scenario on the right
Ask · Acme Payroll
Why does the month-end job fail on 31-day months?
Answer Receipt · click a citation
Answer ReceiptsHonest refusalTenant-scoped memoryCode-linked citationsExpert interviewsArchaeology BriefAnswer ReceiptsHonest refusalTenant-scoped memoryCode-linked citationsExpert interviewsArchaeology Brief
Not another chatbot on your repo.
A serious instrument for teams who cannot afford wrong answers in production.
Answer Receipts
Every sentence links to proof — a line of code, a ticket, a doc section, or a timestamp in an interview.
Honest refusal
No evidence? Backstory says so. Trust is not configurable away.
Expert capture
Code-aware interviews preserve the why before your experts leave.
Linked memory
Git, tickets, and docs indexed per engagement — one tenant boundary, zero cross-leakage.
Trust is the product.
Dual grounding gates, tenant-scoped retrieval, and refusal as a first-class UI state — not an error buried in logs.
0
hallucinated claims shipped
0%
citations on every answer
1
tenant boundary per engagement
∞
institutional memory preserved
How it works
Click a step — same page, no maze of sub-routes.
Step 01
Connect sources
OAuth to GitHub · upload runbooks. Each source shows ingest status in real time.
Before they leave
Stop losing the why when people leave.
One engagement. Connected sources. Receipts on every answer — or an honest refusal.
No source, no claim
Start here
Create an account and index your first sources in minutes.
- Connect Git, tickets, or docs
- Ask with cited Answer Receipts
- Capture expert knowledge before exit