Intelligent Memory OS — Full Feature Reference

Your AI's complete brain. The full toolkit.

No jargon. No architecture diagrams. Just what BizXEngine actually does for your AI — and why it makes your agents dramatically smarter, cheaper, and more reliable.

Start free — no credit card Read the docs →
80%
lower token costs
125×
faster retrieval
60%
less stored noise
99.9%
uptime SLA
AES-256 at rest · TLS 1.3 in transit · 99.9% uptime SLA · GDPR-ready · 15+ PII filters

Why we are unique

Five pillars no other
memory OS combines.

Mem0 ships dedup. Zep ships sessions. Pinecone ships vectors. We ship the complete memory operating system — the only one that does all five at production scale.

Storage Pipeline
7-step intake — PII filter, dedup, scoring, classification — before anything reaches the brain.
Smart Retrieval
pgvector + intent routing + 4-factor hybrid ranking. The right memory, every time.
Self-Optimizing
Health scoring, decay, recommendations, auto-compression. Your brain stays sharp without maintenance.
Plain-language Console
Test, debug, reason — without curl. PMs verify behaviour, engineers debug edge cases.
Production Platform
Workspaces, API keys, rate limits, webhooks, cost caps. Built for teams shipping today.

Core Memory Intelligence

Your AI remembers the right things.
And only the right things.

Memory Storage Pipeline

Every memory is screened before it enters your AI's brain.

When your agent has a conversation, raw messages flow in. Most of that is noise — greetings, duplicates, expired deadlines, sensitive data. We intercept everything before it reaches storage and run it through 7 automatic checks.

The result: only high-signal, clean, categorised memories reach your AI. Nothing embarrassing. Nothing redundant. Nothing that could confuse it later.

60–80%
storage cost saved by filtering noise before it enters
15+
PII filters — passwords, API keys, cards never stored
Live pipeline
Raw conversation input
PII detection — block passwords & cards filtered
Noise filter — remove greetings & small talk filtered
Deduplication — merge near-identical facts merged
Importance scoring 1–10 scored
Temporal classification — persistent or time-bound classified
Stored in pgvector brain ✓ stored
Semantic Deduplication

Your AI stops paying to store the same fact 50 times.

When your agent hears the same information in different words, we don't store duplicates. We detect that "client ships with DHL" and "we always use DHL Express" are the same fact — and merge them into one richer memory.

Before
Client prefers DHL for shipping
We always use DHL Express
DHL Express preferred, avoid UPS
merged →
DHL Express for all intl. shipments. Avoid UPS. imp: 7
40–60% fewer memories stored. Same knowledge, zero redundancy.
Importance Scoring

Not all memories are equal. Your AI finally knows the difference.

Every memory gets a score from 1 to 10 based on its actual business value. Payment terms rank higher than coffee preferences. Critical policies stay at the top. Trivial notes fade away.

Payment terms Net 30 9 / 10
Primary contact: Sarah Lee 7 / 10
Client likes morning calls 4 / 10
Thanks for chatting! 1 / 10
35–50% better retrieval precision. Your AI always surfaces what actually matters.
Temporal Memory & Decay

Old deadlines fade. Timeless policies stay. Automatically.

We classify every memory as either persistent (policies, preferences, contacts — things that don't expire) or temporal (deadlines, offers, seasonal info — things that become stale).

Expired temporal memories don't get deleted — they just fade to near-zero relevance so your AI never confidently cites a deadline that passed six months ago. Important memories decay slowly. Low-value ones fade fast.

Persistent
Payment terms Net 30 — never expires, always relevant
Temporal
Q3 deadline April 30 — fades to 5% relevance after expiry
Decay over time
Important policy (importance 9) 3 years to decay
Meeting preference (importance 5) 6 months to decay
Expired deadline (past date) instant → 5% relevance

Day 0 ————————————————————— Day 1000

Auto-Categorisation

Every memory tagged automatically. Zero manual work.

When you ask "what are our shipping policies?", you shouldn't get pricing facts or random notes. Our categorisation engine tags every memory with its business domain — shipping, payments, compliance, contracts — the moment it's stored.

Ask about shipping → get only shipping memories. 3–5× faster retrieval. Zero noise from unrelated domains.

payment_terms
shipping
compliance
contract
preference
finance
logistics
policy

Auto-detected. No tagging required.

Smart Retrieval

Your AI finds the right memory.
Every time. In milliseconds.

Lightning-Fast Search

Searches millions of memories in under 10 milliseconds.

We use pgvector with HNSW indexing — the same technology powering Notion AI and Supabase. Your agent doesn't wait for memory. It gets the answer instantly, whether you have 1,000 memories or 10 million.

<10ms
retrieval at 1M memories
125×
faster than basic storage
10M+
memories per workspace
Speed at 100K memories
Basic storage (JSON)~1,000ms
BizXEngine pgvector~8ms
Smart Query Routing

Your AI understands what you're actually asking for.

Ask "what changed this week?" and we automatically fetch recent updates. Ask "what are our policies?" and we fetch stable, persistent rules. The retrieval strategy changes with your intent — not just your keywords.

recent "What changed this week?" → fetches last 7 days first
policy "What are our payment terms?" → fetches high-importance persistent rules
exact "What is Client X's account number?" → precision fact lookup
4-Factor Relevance Scoring

The most relevant memory wins — not just the most similar one.

Basic AI memory just picks the closest match. We rank by four signals combined: how similar it is, how recent it is, how often it's been used, and how important it is. The result you get is actually the most useful one.

Semantic match 55%
Recency 20%
Importance 15%
Usage frequency 10%

Self-Optimizing Memory

Your AI's brain cleans itself.
No maintenance required.

Memory Health Score

Know instantly if your AI's memory is in good shape.

Every workspace gets a health score from 0 to 100, updated daily. It factors in how many memories are stale, never accessed, duplicated, or bloated. Score 90+? Your AI is sharp. Score below 60? Time to optimise.

Use it to prove to stakeholders that your AI systems are getting smarter over time — with a number, not a hunch.

90–100: Excellent
75–89: Good
60–74: Needs work
Below 60: Urgent
84
Health Score
Stale memories12
Duplicates detected3
Optimised this week+8
Total active memories2,847
Auto-Recommendation Engine

Your AI's memory fixes itself. You just approve.

Every week, we scan your memory bank for problems — stale facts, bloated entries, duplicates, low-value items that keep getting accessed. We surface them as one-click fixes. Accept, reject, done.

Archive stale
Not accessed in 90+ days → archive
23
Merge duplicates
92%+ similar → merge into one
7
Split overlong
Too long → split into focused chunks
2
Memory Compression

Clusters of similar memories get merged into one rich summary.

When 3 or more memories are saying essentially the same thing, we synthesise them into one comprehensive memory — richer than any single original. The originals are archived. Your AI gets cleaner, denser knowledge.

3 memories → 1 compressed
Client prefers DHL for shipping
We use DHL for international
DHL Express preferred, not UPS
DHL Express for all intl. shipments. Avoid UPS. ↑ imp 7
40–60% storage reduction. Richer context. Better answers.
Memory Decay Engine

A daily background job keeps your AI's memory honest.

Every 24 hours we re-score every memory based on access patterns, age, and importance. Forgotten facts lose weight; frequently retrieved ones gain it. Nothing is hard-deleted — relevance just shifts so your AI stops citing what no longer matters.

Decay runs as a maintenance worker. You see the effect; you never have to schedule it.

Daily decay tick
memories scanned2,847
re-scored upward+182
re-scored downward−96
archived (stale)23
runtime1.4s

Platform

Built for teams, agencies & enterprises.
Production-ready from day one.

Multi-Workspace

One client, one brain. Manage 100 clients from one dashboard.

Each workspace is fully isolated — separate memories, separate billing, separate API keys. Perfect for agencies. Each client's data never touches another's.

API Key Management

Separate keys for dev, staging, and production. Revoke instantly.

SHA-256 hashed, workspace-scoped, expiry-supported. Each key tracks its own usage. One compromised key revoked in seconds — no downtime.

Cost Intelligence

Know your memory spend before the bill arrives.

Real-time usage tracking per workspace and per API key. Projected end-of-month cost. Alerts before you hit limits. Zero surprise bills.

Webhooks & Integrations

React to memory events in real time.

Get webhooks when memories are stored, updated, or when you approach limits. Sync to Salesforce, trigger Slack alerts, log to analytics — all automatically.

memory.stored memory.updated workspace.limit_reached optimization.completed
Security & Compliance

Your AI will never accidentally store a password.

15+ PII filters block passwords, API keys, credit card numbers, and social security numbers before they ever reach storage. AES-256 at rest, TLS 1.3 in transit. GDPR-ready with full data export and deletion.

AES-256
at rest
TLS 1.3
in transit
GDPR
ready
Usage Analytics

See exactly what your AI is doing — and what it's costing.

Per-workspace and per-API-key dashboards: writes, retrieves, reasoning calls, error rates, latency P50/P95/P99. Filter by date, drill into a single key, export to CSV. Health History tracks score trend over 30/90/365 days.

MonTueWedThuFriSatSun
Rate Limiting & Cost Caps

Bursts, runaway loops, and stuck workers can't break the bank.

Token-bucket rate limiting at the workspace and API-key level. Atomic Redis Lua script — sub-millisecond decisions. Hit the cap and the workspace transitions to limit_reached automatically. Reset on the next billing cycle. Optional hard cost ceiling per workspace.

writes / min73 / 100
retrievals / min412 / 1,000
cost cap$84 / $100

Developer Tools

Everything you need to
ship in hours, not weeks.

Memory Console

Test, inspect, and debug your memory — without writing a single line of code.

The Memory Console is a full-featured UI built into every workspace. Store memories, run searches, ask reasoning questions, and see exactly how BizXEngine scores, categorises, and deduplicates each entry — in real time.

Built for both developers and non-technical users. Product managers can verify behaviour. Engineers can debug edge cases. No curl, no Postman required.

Store
Write memories, see importance + category assigned instantly
Retrieve
Search and inspect scores, rankings, and metadata
Reason
Ask questions, get synthesised answers with source IDs
Usage Log
Audit trail of every operation with cost per call
store retrieve reason log
memory text
Client prefers DHL Express for all international shipments. Avoid UPS.
response
status stored_new
importance 9 / 10
category shipping · logistics
temporal persistent
pii_detected none · safe
memory_id mem_9xKzA
Memory Reasoning

Ask your memory a question. Get a real answer — not a list of raw entries.

BizXEngine retrieves the most relevant memories, passes them to an LLM, and returns a clear, synthesised answer. Every answer cites the exact memory IDs it drew from — so you can verify.

POST /v1/memory/reason
query
"What changed this week with Client XYZ?"
reasoning
"Two updates this week: payment terms changed to Net 60, and primary contact switched from John to Sarah."
sources: mem_501 · mem_489
Pro: 1,000 queries/mo Max: 10,000/mo Enterprise: unlimited
Batch Import

Go from empty workspace to 2,000 live memories in under a minute.

Upload a JSON or CSV file — from Salesforce, HubSpot, Notion, or any source. Every entry gets automatically deduped, importance-scored, temporal-classified, and categorised. No manual entry, no custom scripts.

client_data_export.json 92%
1,821
imported
26
dupes skipped
153
remaining
Max 2,000 per import JSON · CSV Auto-enriched on import
REST API · MCP · SDKs

Connect from any language. Any framework. Any agent runtime.

A clean REST surface — store, retrieve, reason — secured by workspace-scoped API keys. The MCP adapter gives Claude Desktop, Cursor, Windsurf, and VS Code memory in one install. SDKs for Node and Python land next.

POST /v1/memory/store POST /v1/memory/retrieve POST /v1/memory/reason
REST · live MCP · live Node SDK · soon Python SDK · soon
curl · store memory
curl https://api.bizxengine.com/v1/memory/store \
-H "Authorization: Bearer $BIZX_API_KEY" \
-H "Content-Type: application/json" \
-d '{"text": "Client prefers DHL Express"}'
{ "memory_id": "mem_9xKzA", "stored": true, "importance": 9 }

How we compare

Built for the things others ship as add-ons.

Most "AI memory" tools cover one slice. BizXEngine ships the full operating system — production-grade, day one.

Capability BizXEngine Mem0 Zep Pinecone Plain pgvector
PII filtering on ingestBuilt-in
Semantic dedup before storage7-stepPartial
Importance scoring 1–10Sessions only
Temporal decay engineDailyManual TTL
Multi-factor hybrid scoring4-factorSimilaritySimilaritySimilarity onlySimilarity only
Intent-aware query routing
Memory health score0–100
Auto-recommendation engineWeekly
LLM reasoning over memoriesBuilt-inDIYAdd-onDIYDIY
No-code Memory Console
MCP adapter (Claude/Cursor/Windsurf)
Workspace-scoped API keysProjectDIY
Token-bucket rate limits + cost capsPer workspaceTier limitsTier limitsTier limitsDIY
Webhook eventsPartialDIY
Real cost intelligenceLive + projectionUsage onlyDIY
Comparison reflects publicly documented features as of 2026-04. Sources: vendor docs & pricing pages.

Ready to give your AI a brain worth trusting?

Start free. Connect in minutes. Your agents will never forget what matters — or remember what doesn't.