Memories Guide

Memories are the raw material of your AI's brain. While Neural Snapshots are structured knowledge crystals, memories capture the everyday facts, decisions, insights, and events that make context real.

What are memories?

Think of memories as your AI's diary. Every fact you tell it, every decision you make, every insight from a meeting — stored as a tagged, searchable record. Unlike snapshots (which are structured thought-crystals), memories are lightweight and fast to create.

Memories vs Snapshots

Snapshots = structured knowledge (16 fields, typed, confidence scores).
Memories = raw knowledge (content + type + tags + optional metadata).

Use memories for quick capture. Use snapshots for deep, structured knowledge. The crystallization pipeline can promote memories into snapshots automatically.

7 Memory Types

Every memory has a type that tells your AI what kind of information it is. This matters for search — when you ask "what did we decide about pricing?" NeuralSnap can prioritize decision memories.

fact

Something objectively true.

"Our ARR crossed $1M in January 2026."

decision

A choice that was made, with rationale.

"We chose Supabase over Firebase because of Postgres + RLS."

insight

A non-obvious realization from experience.

"Users who complete onboarding in <2 min have 3x retention."

story

A narrative with context and emotion.

"The day we almost lost our biggest client because of a deploy bug."

framework

A reusable thinking tool.

"RICE scoring: Reach × Impact × Confidence / Effort."

preference

A personal or team preference.

"Sefy prefers mobile-first reviews before any desktop polish."

event

Something that happened at a point in time.

"Product Hunt launch on Feb 21, 2026 — reached #3."

Creating memories

Use the POST /memories endpoint or the dashboard UI. At minimum, provide content and memory_type.

bash

curl -X POST https://api.neuralsnap.ai/v1/memories \
  -H "Authorization: Bearer $NEURALSNAP_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
  "content": "We chose Supabase over Firebase because Postgres gives us RLS, proper migrations, and we can run complex queries without Cloud Functions.",
  "memory_type": "decision",
  "tags": ["infrastructure", "database", "architecture"],
  "speaker": "Sefy",
  "confidence": 0.95,
  "emotion_tags": ["confident", "excited"]
}'

Response:

json

{
  "id": "mem_xyz789",
  "content": "We chose Supabase over Firebase because Postgres gives us RLS...",
  "memory_type": "decision",
  "tags": ["infrastructure", "database", "architecture"],
  "speaker": "Sefy",
  "confidence": 0.95,
  "emotion_tags": ["confident", "excited"],
  "created_at": "2026-02-21T04:30:00Z"
}

Memory fields

Property	Type	Description
`content`*	string	The memory itself. Be specific — "We decided X because Y" is better than "Discussed database options."
`memory_type`*	string	One of: `fact`, `decision`, `insight`, `story`, `framework`, `preference`, `event`.
`tags`	string[]	Freeform tags for filtering. Keep them lowercase and consistent.
`speaker`	string	Who said or produced this memory. Useful for meeting transcripts.
`confidence`	number	0 to 1. How certain you are about this memory. Defaults to 0.5.
`emotion_tags`	string[]	Emotional context: "frustrated", "excited", "uncertain". Powers the Emotions layer.
`source`	string	Free-text origin reference. E.g. "Q4 planning meeting, Jan 15".

Searching memories

Memories are embedded with the same 768-dimensional vectors as snapshots. You can search semantically or filter by type and tags.

bash

# Find memories about database decisions
curl "https://api.neuralsnap.ai/v1/memories/search?q=why+did+we+pick+our+database" \
  -H "Authorization: Bearer $NEURALSNAP_API_KEY"

Real-world examples

Here's how different teams use memories to give their AI agents persistent context:

🤖 AI Agent Context

Your coding agent forgets what you told it yesterday. With memories, it remembers:

[preference] "Always use pnpm, never npm. We standardized in Jan 2026."

[decision] "API routes use Zod validation. No exceptions. Added after the March incident."

[fact] "The billing webhook endpoint is /api/webhooks/stripe. It handles both checkout.session.completed and invoice.paid."

📋 Meeting Capture

After every meeting, store the key outcomes as memories:

[decision] "Kill the podcast. Reallocate budget to short-form video. Sarah disagreed but team voted 4-1." — speaker: "Team"

[insight] "Users who see the demo video before signup convert 2.3x higher. We need the video above the fold." — speaker: "Pulse"

[event] "Product Hunt launch scheduled for Feb 21. Vera is handling the listing, Echo writes the announcement emails."

📚 Learning & Research

Capture key takeaways from books, podcasts, and articles:

[framework] "Jobs To Be Done: customers don't buy products, they hire them to do a job. Ask: what job is the customer hiring this for?"

[insight] "From 'Thinking Fast and Slow': anchoring bias means the first number people see dominates their judgment. Show the higher price tier first."

Best practices

✅

Be specific

"We chose Supabase over Firebase because of RLS and Postgres" > "Discussed database options"

✅

Use the right type

A decision is not a fact. An insight is not a preference. Correct typing improves search relevance.

✅

Tag consistently

Use lowercase, hyphenated tags. Pick a vocabulary and stick to it: "content-strategy" not sometimes "content" and sometimes "marketing-content".

✅

Include the "why"

"We switched to weekly deploys because bi-weekly was too slow for hotfixes" — the rationale is as valuable as the decision itself.

✅

Attribute speakers

Set the speaker field. Six months from now, "who said we should kill the podcast?" becomes a searchable question.

From memories to snapshots

Memories are the input; snapshots are the refined output. When you accumulate enough related memories, the crystallization pipeline can synthesize them into structured snapshots.

For example, 5 memories about "short-form video working better than long-form" might crystallize into a single Conviction snapshot: "Short-form video is the highest-leverage content format at our current scale."

Auto-crystallization

Connect Fathom or Fireflies to automatically store meeting memories and crystallize them into snapshots — zero manual work.

Concepts

Crystallization Guide