About
Built so changed facts stay inspectable
MemStrata is built by Neeraj Yadav. The public products help you keep useful context, inspect sources, and work with the reader you choose.
The problem that started this
Useful work accumulates in conversations, repositories and operational records. When a fact changes, retrieval often still surfaces the earlier statement as if it were current. I wanted a system that keeps the earlier statement openable and shows the supporting record, rather than asking people to trust an uninspectable summary.
Why four products
Conversational is the starting product: bring your own material, connect the AI you already use, then inspect the evidence. Coder supplies selected repository context to a supported coding tool. Structured answers supported current, history and change questions over mapped records without requiring a generative model. Governance observes connected agent actions and records the policy decisions you configure.
What is available today
Four installable products on Windows x64, Linux x86_64 and macOS 15 Apple silicon. Conversational current wheel is 0.7.19. Each product has a 180-day trial after activation. Account services and selected bounded policy decisions use MemStrata services. A cloud reader receives only the content needed for the connection you choose.
MemStrata is an independent early-stage product. We are looking for design partners and integration pilots, not a large published customer roster.
Press kit
One sentence: MemStrata gives AI a memory you can check.
One paragraph: MemStrata keeps chats, documents, code history and records in an index on your computer, answers with the model you choose, and shows the sources behind an answer, including when a fact has changed.
Who this is for
Design partners who will install one product, bring a real source or repository, and report whether the first inspectable answer is useful. Integration pilots who want evidence retrieval beside an existing application, without replacing that application. Research readers of the 4 published arXiv preprints in the research registry.
Founder and research
- Founder Neeraj Yadav
- Paper 1 Temporal Validity in Retrieval Memory: Eliminating Stale-Fact Errors for AI Agents over Evolving Knowledge
- Paper 2 Temporal Validity on Real Software Histories: Eliminating Stale-Fact Errors in Code-Assistant Memory over GitHub Fixes
- Paper 3 When Old Facts Return: Re-Reads, Reverts, and the Limits of Temporal Memory
- Paper 4 MemStrata: 95% and 90.91% Source-Aware Accuracy on LongMemEval-500 and LoCoMo-1540 with a Local Qwen 3.8 27B Q4_K_M Reader
- Python packages PyPI
- LinkedIn MemStrata company page
Contact
- Design partner or integration pilot Discuss your use case
- General hello@memstrata.dev
- Support support@memstrata.dev
- Security security@memstrata.dev