Intelligence · Memory

A product memory layer for every decision.

Every outcome — shipped, parked, deferred — is recorded. When you start a new brief, annsa surfaces what you shipped in the same area before, so the new work stands on the old.

annsa
Memory

What annsa remembers about co-edit

Tandem·9 outcomes recorded·since Jan
Past shipsWhat you closed here, and how it landed.
  • Co-edit conflict fix
    Shipped Apr · 14 people notified · no regressions since
    co-edit.ts
  • Presence cursors for shared blocks
    Shipped Feb · 6 accounts asked · still the most-used view
    Canvas.tsx
  • AI thread summary
    Deferred to next quarter · reason and history kept
CorrectionsThe changes you made that annsa carried forward.
  • You renamed “Realtime” → “Collaboration”
    Re-clustered 31 signals · the new name carries into briefs
  • You re-ranked block locking above export polish
    Annsa weights co-edit work higher here now
Related contextWhat bit last time you touched this ground.
  • Locking the block briefly broke presence cursors
    Last co-edit change · Simon caught it · watch this path
    usePresence.ts
  • Ben flagged merge order on queued edits
    Replay the queue after the rewrite lands, not before
    co-edit.ts
  • Catherine asked for a shipped-note on this one
    Close the loop with the 14 who reported the conflict
— The problem —

Every project starts from zero, even when it shouldn’t.

You solved a related problem eight months ago. You learned, the hard way, where the billing work tripped you up last time. But that knowledge lives in your head, or in a closed ticket no one will reopen, so the next brief in the same area starts blank — and the team rediscovers what it already knew, at full cost, again.

Annsa remembers. Every shipped, parked and deferred outcome is recorded, and when a new brief touches the same ground, the prior outcome surfaces before you start drafting.

— How it works —

From outcome to memory to the next brief.

Memory is written when work closes and read when new work begins.

01
Capture the outcome.

When a spec ships or a priority is parked or deferred, annsa writes a memory row — what shipped, what the outcome was, what corrections were made.

tandem.app · spec closed
Co-edit conflict fixShipped
AI thread summaryDeferred
Legacy comment importKilled
02
Store it for relevance.

Each memory is embedded so it can be found by meaning, not by remembering it exists.

Embedded · found by meaning
“Co-edit conflict fix”
vector · 1,536 dims
Indexed for retrieval
Searchable by similarity
03
Surface it on the next brief.

When a new brief touches an adjacent area, get_relevant_memory pulls the prior outcome before you start drafting.

get_relevant_memory
New brief · “Block locking in co-edit”
Match · 0.89 similarity
You shipped “co-edit conflict fix”
Surfaced before drafting
04
Tune the strength.

Pick Conservative, Balanced or Aggressive for how strongly your corrections steer future briefs.

Memory aggressiveness
ConservativeLight
BalancedDefault
AggressiveStrong
— What makes it different —

The system learns from what happened, not just what came in.

Most tools learn from inputs — more feedback, more tags. annsa also learns from outcomes — what you shipped, what you parked, what you corrected. That’s the difference between a record and a memory.

Context from what happened before.

When a new brief touches an area you’ve worked in, annsa surfaces a reference to what you shipped there last. “You shipped ‘export polish’ previously” — before you start, not after.

Risks from evidence, not guesswork.

The risk section of a brief is populated from past outcomes and signal patterns in adjacent areas, instead of invented caveats.

Tunable aggressiveness.

Pick Conservative, Balanced or Aggressive. The control sets how strongly your corrections — renamed themes, re-ranks, refinements — steer future briefs.

— Everything in this feature —

The rest of what compounds.

Memory aggressiveness control

Three levels: Conservative, Balanced, Aggressive.

Correction memory

Rename a theme or re-rank a priority once and the change carries into future work.

Win/loss tagging

An honest record of what shipped, what was parked and what was deferred, kept against each brief.

Brief-time memory surfacing

Relevant past outcomes appear before you start drafting, retrieved by embedding similarity.

Priority-time “we’ve touched this before” nudge [planned]

Today the nudge appears at brief-draft time, not when the priority forms.

Outcome cohorts [planned]

Compare how different kinds of briefs performed over time.

Patterns across what ships [planned]

Surface what kinds of work succeed once enough briefs have shipped to compare.

— Questions —

Asked and answered.

What does annsa remember?

Outcomes — what shipped, what was parked, what was deferred — along with the corrections you make and the themes you rename. The record of how your product decisions played out.

How does memory show up in my work?

When you start a brief in an area you’ve worked in before, the relevant past outcome surfaces before you start drafting, retrieved by embedding similarity.

Where do the risks in a brief come from?

From past outcomes and signal patterns in adjacent areas — the risks you’ve encountered — rather than generic caveats.

Does correcting annsa teach it?

Yes. Renaming a theme or re-ranking a priority is remembered and applied to future briefs through the memory aggressiveness control.

Can I control how much memory influences new work?

Yes. Memory aggressiveness is tunable — Conservative leans on your corrections lightly, Aggressive surfaces them strongly. You set how much weight your past corrections carry.

Does this get more useful over time?

That’s the point. Every outcome adds to what annsa knows. A team running it for a year is working with a system that has learned from everything they shipped.

Go deeper

Learn more about the annsa Intelligence suite.

Stand on every decision you’ve already made.

Memory is part of annsa’s Intelligence suite — where autonomous product intelligence compounds.