About · DataFrontier Innovations
Built as research.
Shipped as product.
Smritium is developed by DataFrontier Innovations — a research-led AI lab in Bengaluru building the sovereign memory layer for AI: the private, portable context substrate (UPII) for individuals and its governed counterpart (UEII) for enterprises. We publish what we learn, and productise what compounds.
The thesis
Models are no longer the hard part.
The hard problems now live in everything around the model — memory, identity, and the substrate between a person (or an enterprise) and the agents that act for them. That is where we work.
Our core claim: personal and institutional context is a primitive, not a feature — a typed, addressable, permissioned store of who you are, what you've done, and what you're working on, that any model can bind to with consent, the way applications bind to a filesystem rather than reinvent one.
Smritium is the first product expression of that research: a cross-vendor, local-first memory substrate that tests the thesis with a single sitting of real user experience.
What we build
Two substrates. One memory layer.
Unified Personal Intelligence Infrastructure
A private, sovereign memory substrate for individuals — a typed, addressable, permissioned context graph any model can bind to, with consent, that lives on the user's own device.
Unified Enterprise Intelligence Infrastructure
The same substrate, governed for organisations — role-based access, data lineage, and cross-functional learning loops that turn one team's insight into shared institutional memory.
The local memory substrate
The engine beneath both: content-addressed capture, multi-tier storage, and hybrid retrieval — sovereign by construction, verifiable, and offline by default.
The research, in depth
From stateless agents to a sovereign memory substrate.
The complete argument — the problem with stateless AI, the UPII concept, the local-first architecture, and the empirical evidence behind it.
10.7×
fewer prompt tokens vs. replaying flat conversation history
87.5%
answer accuracy on the LoCoMo long-memory benchmark
100%
Hit@10 retrieval on LoCoMo
181%
3-year ROI from a persistent enterprise memory layer
The fundamental limitation of today's AI agents is structural statelessness. Large language models have no native mechanism to persist information across separate API calls, so any context generated in a session lives only inside the active context window. When the session ends or the window fills, the agent suffers immediate context collapse.
For users this means constant repetition, personalization that can't survive long horizons, and agents that forget past workflows and preferences. Developers paper over it by replaying raw, uncompacted transcripts into the prompt — which creates a steep linear cost curve, adds latency, and degrades the model's attention as it hunts for real instructions inside bloated context.
Worse, because people use many independent AI clients — browser copilots, IDE assistants, chat apps — personal context gets trapped inside disconnected, proprietary databases. The result is conflicting histories, agents that can't share reasoning, and hard vendor lock-in: you can't port your interaction history between tools.
Cutting-edge — and measured
Claims we can reproduce, not just assert.
Provisional patent
On multi-signal context rehydration — fusing semantic, temporal and relational signals into one ranked context window.
Reproducible by design
Re-ingesting the same corpus reproduces 100% of chunk hashes. Citations stay stable across re-indexes and re-embeddings.
Benchmarked retrieval
Recall@10 = 0.958 on a committed labelled set — with a public harness so anyone can reproduce the number.
Six open questions
From the right unit of memory promotion to provenance under inference — the questions that decide whether the substrate is real.
Government of Karnataka
Backed by the ELEVATE Nxt programme · 2026
A deep-tech grant-in-aid initiative supporting frontier AI research built in India.
Smritium is developed under the ELEVATE Nxt programme, Government of Karnataka — a deep-tech grant-in-aid initiative supporting frontier research built in India.