Memory that respects what users actually said.
Where TensorPRO operates a chat or demo experience directly with people, such as EDN Chat, we inform you clearly, and at the point of interaction, that you are interacting with an AI system rather than a person. EDN itself is model-agnostic middleware that runs in the background, you connect your own LLM. Where a customer builds their own product on EDN, responsibility for that product's own AI-disclosure rests with the customer as its operator.
Every memory record EDN stores is classified by where it came from at the point it is written, whether a user stated it, a user confirmed it, a document verified it, the AI inferred it, or the AI generated it. Records the AI generated without a verified source are structurally ineligible to be recalled as fact. This is enforced in the data model, not by a confidence score, and it cannot be overridden by recency or relevance.
When a user corrects something the system captured, that correction is stored as its own record, distinct from the original, rather than silently editing history. Automatically reconciling a correction with the exact record it supersedes is on our near-term roadmap.
We do not use the content you submit to train models for other customers, or to build competing products. The full commitment is in our Terms of Service.
Every record carries its provenance, timestamp, and originating session. There are no hidden embeddings silently shaping behaviour, an operator can inspect exactly what EDN holds and where each item came from.
These commitments are supported by our published evaluation. See Research for methodology and results.
Memory infrastructure for AI systems. No fine-tuning. No model changes.
Integration guides, API reference, and architecture documentation for developers.
View documentationBenchmark results, methodology, and the provenance model behind EDN Memory Engine.
View research