Mnexium AI

Mnexium is built to drop into an existing app instead of replacing it. You add an mnx object to your cURL or HTTP calls, or use the official SDKs for Node.js and Python. That's it. That small change turns on long-term memory, so your LLM app remembers users across sessions without a custom vector database.

The platform separates context into chat history, agent memory, agent state, and records. Each runs independently. It's a context management API where you can keep session-level continuity without touching long-term memory, or store task state without polluting user facts. Keeping them apart is what lets the same API serve both personalized chat and deterministic data workflows.

Mnexium routes to OpenAI's Chat Completions and Responses APIs, Anthropic, and Google's Gemini without locking you to one provider. Memories and records stay accessible across providers when you keep the same subject_id. Switch models whenever you like. Your agent's memory doesn't reset.

With learn: true set, the platform pulls facts and preferences out of conversations on its own. Records go further: set records.learn to auto or force and Mnexium extracts structured entities from the dialogue. That means you get clean data without writing extraction prompts yourself.

Every request goes through a fixed assembly order: system prompt, agent state, memories, records, chat history, then your user message. Because the order is predictable, you can reason about exactly what the model sees. Logs, traces, and recall events show what was injected and why, which makes debugging far easier than guessing.

Mnexium exposes logs and recall events for every step. You can inspect which memories were retrieved, which records were updated, and what prompt the model received. For teams shipping an AI memory feature, that visibility is the difference between shipping and shipping blind.