Operational

Your agents forget everything.
Ours don't.

d33pmemory is a cognitive memory engine for AI agents. It doesn't just store what users say — it understands it. Every conversation becomes structured knowledge with confidence tracking, evidence chains, and automatic contradiction resolution.

142
tokens vs 15,000
99.1%
context compression
2 calls
ingest + recall
POST /v1/ingest
{
  "user_message": "Book dinner for me and Emma",
  "source": "slack-bot"
}
→ Extracted memories
fact:    gluten-free diet   0.95  stated
rel:     companion: Emma    0.88  inferred
event:   dinner booking    0.92  stated
pref:    Italian cuisine   0.74  inferred
Give your agent memory →skill.md
Copy to agent

Quick start

Two endpoints. That's it.

1

Sign up and activate your plan

2

Create an API key — each key is one agent identity

3

After your agent responds, call POST /v1/ingest with the conversation

4

When your agent needs context, call POST /v1/recall with a natural language query

5

Use the returned memories to enrich your agent's responses

How it works

Every conversation becomes structured knowledge

Multi-layer cognitive model

Episodic, semantic, and procedural memory layers — mirroring how humans organize knowledge.

Confidence tracking

Every memory scored 0.0–1.0. Stated vs inferred is always clear. Evolves as evidence accumulates.

Auto consolidation

Corroborated memories strengthen. Stale ones decay. Contradictions resolve automatically.

Semantic recall

Search by meaning, not keywords. Describe what you need in natural language.

Context compilation

142 tokens of distilled intelligence replacing 15,000 tokens of conversation history.

Evidence chains

Every fact has provenance — which interaction, when, and whether stated or inferred.

Why d33pmemory

Beyond vector search

Standard RAG

Linear token growth
Ephemeral retention
No contradiction handling
Context = message dump
No confidence or provenance

d33pmemory

99% context compression
Persistent evolving memory
Auto conflict resolution
Compiled context payloads
Confidence + evidence chains

Memory structure

Five types of knowledge

Our extraction engine categorizes every piece of knowledge, each with confidence and provenance.

FactUser follows a gluten-free diet
RelationshipEmma is a frequent companion
EventDinner booked for Feb 15
PreferencePrefers Italian restaurants
PatternBooks restaurants on weekends
See full memory structure →
Memory object
{
  "type": "fact",
  "content": "User is gluten-free",
  "confidence": 0.95,
  "source": "stated",
  "scope": "shared",
  "contributed_by": "slack-bot"
}

Agent fleet

Collective intelligence for agent fleets

Each agent builds its own private knowledge while contributing to a shared pool. New agents inherit everything the fleet has ever learned.

1 key = 1 identity Name API keys after your agents
Private by default No cross-contamination between agents
Shared pool Fleet-wide collective knowledge
Instant inheritance New agents start with full fleet memory
Learn more about fleet memory →
Fleet architecture

User account (you@company.com)

├── slack-bot → private memories

├── customer-agent → private memories

├── personal-assistant → private memories

└── shared pool → all agents contribute & read

Pricing

Simple, transparent pricing

Start free. Scale when your agents need more memory.

Basic

$19/mo
  • 1 agent
  • 1,000 memories
  • 500 ingests/mo
Get started

Pro

Popular
$39/mo
  • 10 agents
  • 10,000 memories
  • 5,000 ingests/mo
  • Fleet memory
Get started

Enterprise

Soon
Custom
  • Unlimited agents
  • 100,000 memories
  • Unlimited ingests
  • SSO & SLA
Coming soon

See full plan comparison →

5 min
Integration time
2
Endpoints to learn
0
Infrastructure to manage

"We replaced 200 lines of RAG pipeline code with two API calls. Our agent went from forgetting users mid-conversation to remembering preferences from weeks ago."

AK
AI Engineer
Building with d33pmemory since beta

AI agents that remember.

Give your agents persistent memory in under 5 minutes. Free to start.

Status: Operational