AI lets teams move faster than their memory can keep up.
Assay is a memory layer for Claude Code. It keeps track of what was decided, why, what changed, and what still holds, and brings it back in later sessions, so work that ships at AI speed stays coherent.
Search retrieves artifacts. Assay resolves decision state.
Why this exists now.
AI did not just make teams faster. It changed where the bottleneck is.
AI output compounds.
Human context does not.
Agents produce specs, code, analysis, and options faster every cycle, and that output is durable. The reasoning behind your decisions still lives in memory, meetings, and docs that go stale.
When output outruns context,
teams drift.
You can build more than you can keep coherent. Settled questions get reopened, work ships against assumptions that already changed, and nobody is sure which decision is the current one.
The new bottleneck is
keeping decisions current.
The hard part stops being “can we build it” and becomes “does this still match what we decided.” That gap is what Assay is built to close.
What Assay does.
A memory layer on top of the notes and transcripts you already have, adding what none of them track: which decisions still hold. Not more stored text. A record of what holds.
You found the doc.
You still don’t know if the decision stands.
Assay records the current answer, the reasoning behind it, and the decision that replaced it.
Two months later, the team remembers
the argument, not the outcome.
Decisions are captured as typed records with rationale, owners, and context, not reconstructed from Slack archaeology.
Roadmaps drift when old calls
quietly get reopened.
Assay keeps a clear trail of what was reversed and what replaced it, so stale decisions stop resurfacing as live ones.
How you use it.
Assay lives inside Claude Code as a small set of commands. Two do most of the work. Code and full docs are on GitHub.
ALSO INCLUDES
The questions worth answering first.
If you’re reading this cold, start here.
How is this different from Notion, search, or ChatGPT memory?
Those tools store and retrieve text. Assay tracks decisions: what was decided, why, and what superseded it. It sits on top of whatever memory you already use and adds the decision layer none of them have. You don’t just find notes; you resolve the current state.
Why does AI make this matter more?
Because AI raises the volume of specs, code, and options a team produces, but the context behind your decisions still decays at human speed. The faster you ship, the faster settled decisions go stale without anyone noticing.
What do I actually use day to day?
Start with two commands. /assay-decision brings back your recent decisions, each linked to the session where it was made. /assay-brief gives you a short, cited brief on a topic. To save something on purpose, use /assay-decision-deposit. Decisions get captured as you work, so you don’t need a save command.
What stage is this at?
Early and honest about it. Assay is a working local-first prototype inside Claude Code, single-user. Everything lives in one file on your laptop, so your decisions never leave your machine. We’re explaining it openly while we learn where it’s most useful.
Try the prototype.
For people who want to run it inside Claude Code today. Local-first and single-user.
# inside Claude Code > /plugin marketplace add levievanshantz/assay > /plugin install assay@assaylabs # restart Claude Code, then > /assay-decision > /assay-brief onboarding ✓ local decision memory, ready inside Claude