Einstein — our own memory system that learns from every conversation
A unique long-term memory architecture built for AiHummer. Your AI employees accumulate knowledge from every dialogue and bring it into every reply — locally, on your server, with no external paid services.
What makes it different
Learns from every conversation
Long-term memory is not a chat log: facts, decisions and preferences are captured from every dialogue and become part of the agent’s knowledge.
Semantic + episodic recall
The agent finds knowledge by meaning, not keywords — and can also recall specific past episodes with stable citations back to the source.
Structured fact extraction
Candidate facts are extracted from conversations into a dedicated store and go through verification — the memory corpus is never silently rewritten.
“Dream” consolidation
In the background Einstein consolidates what was learned — like sleep for a human: it links related facts, strengthens the important and prunes the noise.
Ambient memory in every turn
Relevant memories are brought into the context of every reply automatically — the agent remembers you without being asked to “search its memory”.
Local embedder, no paid APIs
Vector search runs on a local multilingual embedder (multilingual-e5) shipped with the platform: no external embedding services, no per-request fees, and your data never leaves the server.
Transparent source of truth
The memory corpus is canonical, human-readable Markdown. You can read exactly what the agent knows, review candidate facts and audit every change.
Web UI: graph, review, audit
A protected memory console: browse the memory, explore the knowledge graph, review extracted facts, check quality gates and the audit log.
How it works
An employee or customer talks to the agent in any channel.
Einstein extracts candidate facts and episodes from the dialogue.
Verified knowledge lands in the canonical corpus; background consolidation links and strengthens it.
Relevant memories are retrieved semantically and injected into the agent’s context on every turn.
In the base install — always
Einstein is bundled with every AiHummer installation — a non-removable part of the platform. Nothing to buy, connect or configure: memory works out of the box.
Questions
Is Einstein a paid add-on?
No. Einstein ships in the base AiHummer install on every plan, including the free Community tier, and cannot be removed — memory is a core capability of the platform.
Does it need external services?
No. Retrieval, embeddings and consolidation run locally on your server (host-native, systemd, no Docker). No external paid memory or embedding APIs are used.
Can I see what the agent remembers?
Yes. The memory corpus is human-readable Markdown, and the protected Web UI shows the memory overview, the knowledge graph, fact review queues and the audit log.