AI MEMORY
IN 2 LINES OF CODE
Add persistent memory, chat history, user profiles, records, and live context to your AI app without building memory infrastructure yourself.
Try the live demo to watch memory persist across sessions. Works with OpenAI, Anthropic, and Gemini.
Evaluating memory infrastructure? See how the AI memory API works.
Type a fact about yourself and open the demo again to see memory persist across sessions.
import { Mnexium } from "@mnexium/sdk";
const mnx = new Mnexium({
openai: { apiKey: process.env.OPENAI_KEY },
model: "gpt-4o-mini",
learn: true, // Extract memories
recall: true, // Inject context
history: true, // Prepend chat
});
const alice = mnx.subject("user_123");
const res = await alice.process("My name is Alex");See the difference
Same API. Same models. But now your AI remembers.
Request 1
{ "messages": [{ "role": "user", "content": "I'm allergic to peanuts and vegetarian" }] }Request 2 — new session
{ "messages": [{ "role": "user", "content": "Suggest dinner recipes" }] }Response
“Here are some recipes: Chicken stir-fry, Pad Thai with shrimp…”
Recipes contain meat & peanuts
Request 1
{ "messages": [{ "role": "user", "content": "I'm allergic to peanuts and vegetarian" }],
"mnx": { "learn": true } }Request 2 — new session
{ "messages": [{ "role": "user", "content": "Suggest dinner recipes" }],
"mnx": { "recall": true } }Response
“Since you’re vegetarian and allergic to peanuts, here are safe recipes: Mushroom risotto, Veggie pad thai with cashews…”
Your AI Doesn’t Remember.
Now It Can.
Mnexium adds memory and context across sessions, users, and models.
Your product layer
Your Apps & Agents
Mnexium sits between your app and model, adding memory and context to every request.
Mnexium
Give Your AI a Memory
Memory
Learns facts, preferences & context across sessions.
Context
Injects history, records & live data directly to your AI.
Works with your model stack
Same memory layer across all models
ChatGPT
Claude
Gemini
Drop-in integration
Add Memory to Any AI in 2 Lines
Your AI stays the same. It just starts remembering.
- Learns automatically.
- Extracts facts, preferences, and context from every conversation.
- Injects the right context.
- Surfaces relevant memories on every request.
- Keeps context clean.
- Ranks, deduplicates, and scores what matters.
That’s it. Your AI now has memory.
// Everything stays the same
const client = new OpenAI({
apiKey: MNX_KEY,
baseURL: "https://mnexium.com/api/v1"
})
const res = await client.chat.completions.create({
model: "gpt-4o-mini", // or Claude, Gemini, etc.
messages,
mnx: { learn: true, recall: true }
})Advanced usage
# Search memories by semantic similarity
curl -G "mnexium.com/api/v1/memories/search" \
-H "x-mnexium-key: $MNX_KEY" \
--data-urlencode "subject_id=user_123" \
--data-urlencode "q=favorite food"// Search memories by semantic similarity
const alice = mnx.subject("user_123");
const results = await alice.memories.search("favorite food");# Search memories by semantic similarity
alice = mnx.subject("user_123")
results = alice.memories.search("favorite food")Start free, scale as you grow
Free
$0
Build and test AI apps with persistent memory.
Up to ~100 users
10,000 / month
1,000 / month
- AI apps that remember users across sessions
- API access (no signup required)
- Community support
Builder
$29/month
Launch production AI apps that remember thousands of users
~3,000-10,000 users/month
250,000 / month
7,500 / month
- $0.22 per 1k extra requests
- $3.00 per 1k extra memory ops
- Email support
Growth
$149/month
Designed for scaling apps with consistent traffic
~50k-200k users/month
2,000,000 / month
100,000 / month
- $0.15 per 1k extra requests
- $2.00 per 1k extra memory ops
- Priority support
Enterprise
Custom
For high-volume teams with custom requirements
Unlimited scale
Custom
Custom
- Custom deployment models
- Priority SLA and onboarding support
- Optional contract + invoicing
2 lines of code · Drop-in integration
More than Memory
Mnexium gives your AI memory, state, integrations, and structured context, everything it needs to build real applications.
Why teams choose Mnexium
| Mnexium | DIY Stack | Vector DB | LLM | |
|---|---|---|---|---|
| Time to build | Minutes | Weeks+ | Days | None |
| Completeness | Full | Partial | Partial | None |
Memory | ||||
| Persistent memory | ✓ | ✓ | ✗ | ✗ |
| Fact extraction | ✓ | ✗ | ✗ | ✗ |
| Semantic recall | ✓ | ✗ | ✓ | ✗ |
Application Layer | ||||
| Records (CRUD + search) | ✓ | ✓ | ✗ | ✗ |
| Agent state | ✓ | ✓ | ✗ | ✗ |
| Chat continuity | ✓ | ✗ | ✗ | ✗ |
Platform | ||||
| Managed integrations | ✓ | ✗ | ✗ | ✗ |
| Works with all models | ✓ | ✓ | ✓ | ✓ |
| One API | ✓ | ✗ | ✗ | ✗ |
| No infra to manage | ✓ | ✗ | ✗ | ✓ |
Production AI needs a context layer, not another prompt trick.
Every AI product eventually rebuilds memory, history, user profiles, business object lookup, and context assembly. Mnexium turns that repeated glue into shared infrastructure.
Application burden
AI apps are stateless by default
Models do not remember users, conversations, business objects, or live operational state unless the application rebuilds that context on every request.
Who is this user?
What happened before?
What are their preferences?
What should the model see now?
Runtime boundary
Context needs a runtime
Mnexium sits between the application and the model, assembling the right user memory, business records, history, and constraints before every request.
Application
Mnexium runtime
Model(s)
Control plane
Context becomes the control plane for AI apps
As AI moves from chat demos to production workflows, applications need a governed layer for memory, records, policies, and live context assembly.
Infrastructure layer
A governed context plane for agents
Platform properties
Built as infrastructure
Mnexium is designed as a durable runtime layer, not prompt glue. The white paper documents how memory, truth, profiles, records, integrations, and context assembly work together.
- Separated layers
- Memory, history, profiles, records, and integrations are modeled as distinct systems.
- Governed runtime
- Context can be assembled, constrained, recalled, and audited before each model request.
- Model agnostic
- Context and memories can move across models, so applications are not locked to a single provider.
- Expandable surface
- The same layer can support assistants, agents, copilots, and workflow automation.
Frequently asked questions
How do I integrate Mnexium with my existing code?
Install our Node.js SDK (@mnexium/sdk) or Python SDK (mnexium), or just add an `mnx` object to your existing cURL/HTTP calls. We support OpenAI's Chat Completions and Responses APIs across all approaches.
Do you have official SDKs?
Yes! We offer official SDKs for Node.js (npm install @mnexium/sdk) and Python (pip install mnexium). Both provide a simple, idiomatic interface — initialize the client, create a subject, and call process(). Configuration like learn, recall, and history can be set once at the client level.
What is a memory and how does versioning work?
A memory is a structured record. When new memories conflict with existing ones, the old memory is automatically marked as 'superseded' while the new one becomes 'active'. You can visualize the full evolution chain with Memory Graphs.
What's the difference between Memory, Agent State, and Chat History?
Chat History is the raw conversation log within a session. Agent Memory stores long-term facts about users that persist across all sessions. Agent State is short-term, task-scoped for tracking agent progress and pending actions.
How is Mnexium different from a vector database?
Mnexium is an opinionated memory layer built for AI agents. We handle automatic learning, semantic deduplication, memory versioning, recall scoring, user profiles, governance, and full observability—not just embeddings.
Can I use my own API keys?
Yes. Pass your OpenAI key via x-openai-key, Anthropic key via x-anthropic-key, or Google key via x-google-key. We never store your provider keys—they're used only for the duration of the request.
What models and providers do you support?
We're provider-agnostic. Currently supported: OpenAI ChatGPT, Anthropic Claude, and Google Gemini. Use what you're already used to.
Is my data secure?
Yes. All data is encrypted at rest and in transit. We support scoped API keys with granular permissions (read/write/delete per resource type), and you can revoke access instantly. Full audit logs track every action.
What data do you store?
Mnexium stores the memory, history, profiles, records, and related metadata needed to power your application features. Provider API keys passed on requests are used only for the duration of the request and are not stored.
How do you handle security and retention?
Data is encrypted in transit and at rest, access is controlled with scoped API keys, and audit logs help you track activity across your workspace. Retention and lifecycle behavior depend on the resource and configuration you choose, so you stay in control of what persists.
What are User Profiles?
Profiles are business defined and AI-generated summaries of a user. They provide a overview of preferences, facts, and context—useful for personalizing responses without loading all individual memories.
Want product news and updates? Sign up for our newsletter.
Contact Us
Have questions about Mnexium? Want to discuss integrations?
We'd love to hear from you.
Ready to give your AI a memory?
Start building with persistent memory today. Free tier includes everything you need to get started.
Want the technical architecture? Read the white paper