Memory of every decision your team makes about its tests.
testrelic memoryEvery test run becomes shared context for Cursor, Claude Code, Copilot, and Codex — over MCP. So your whole team ships at senior-IC speed, not just the engineer who's been here longest.
Captured from every surface · 14 engineers, one memory
The journey of one fact
selector.checkout.pay
[data-test=pay-now] → [data-test=pay-submit]
—
waiting for the first capture…
Every capture above becomes a fact the layer can serve — and, eventually, retire.
LevichHow Memory CORE works
The agent proposes — you approve — the whole team remembers
Decide in Ask AI
Work through a test decision in any repo-scoped Ask AI chat — quarantine a flaky spec, agree on an assertion convention, note a refactor.
▌checkout.spec.ts keeps failing on firefox — should we quarantine it?
✻ Ask AI
It isn't timing. [data-test=pay-now] was renamed in v2.4.1 — all 4 failures are post-deploy.
▌agreed — fix the selector, don't quarantine
a decision has been made — nothing saved yet
Approve the save
When you agree, the agent proposes a memory. An Approve / Reject prompt appears above the composer — nothing is written until you confirm.
⧉ Save this as a memory?
Decision
Fix [data-test=pay-now] → [data-test=pay-submit]. Do not quarantine checkout.spec.ts.
proposed · auto-categorized · @sarah
nothing is written until you confirm
Remembered everywhere
It's saved, auto-categorized, and folded into the repo memory digest — read by Ask AI on every future turn, by the team, and by your agent over MCP.
✓ saved · Decision · checkout.spec.ts
folded into repo digest · 148 memories
read back on every turn by
Ask AI · digest injected
TestRelic CLI · /memory
Your agent over MCP · Beta
visible to · the whole team · 14
Four kinds of memory
Every memory is auto-classified
So the right context surfaces at the right time.
Decision
Agreed verdicts — quarantine this spec, keep that assertion, delete the dead test. The calls your team has already made, so they aren't re-debated.
Decision
Quarantine cart.spec.ts until v2.4.1 ships.
@sarah · Mar 12
Decision
Assert on the psp:receipt event, never a fixed wait.
@dana · Mar 09
Decision
Delete the legacy guest-checkout spec — flow removed.
@amir · Feb 27
Insight
Observations about the suite — what flakes, why a flow is slow, what a failure really meant. The knowledge that usually lives in one engineer's head.
Insight
checkout.spec.ts fails 4/30 — all post-deploy, not timing.
from run #4821
Insight
Checkout P75 went 3.2s → 8.4s after the PSP swap.
v2.4.0 → v2.4.1
Insight
Firefox shard is 2× slower than chromium on this suite.
shard 2/4
Maintenance
Follow-up TODOs and refactoring notes — what to revisit, what to clean up, what's deferred. Durable, not lost in a chat scrollback.
Maintenance
Revisit the retry wrapper on login.spec.ts next sprint.
deferred · Apr
Maintenance
Split checkout.spec.ts — 14 tests in one file.
@sarah · TODO
Maintenance
Drop the old data-test shims once v2.5 lands.
blocked
Context
Constraints and conventions — the always / never rules. So neither a teammate nor an AI agent re-breaks them.
Context
Never add waitForTimeout in checkout. Always await the receipt.
always / never
Context
Prefer [data-test] over CSS or text selectors.
convention
Context
Staging PSP is rate-limited — don't parallelise past 4 workers.
constraint
Stop re-litigating your test decisions.
Memory CORE is included in the Growth plan. Start your 14-day free trial — no credit card required. Manage it from the repo Memory tab, the TestRelic CLI, or your coding agent over MCP.
Not to be confused with Ask AI's conversation memory — Memory CORE is per-repository, team-shared, and built from your test decisions.