Evidence memory
Files, session records, public documents, and selected corpora become searchable memory with keyword search, embeddings, reranking, and compiled facts.
Contact
Bring the part of AI that cannot be allowed to guess.
Remanentia is for systems where an answer has consequences: private knowledge, customer-facing retrieval, regulated records, research memory, or operational decisions that need evidence instead of plausible text. A useful first message says what the system must remember, what it must never expose, and where the current stack fails.
Why this exists
LLMs are useful, but memory and trust cannot be left to the prompt.
A model can speak fluently while forgetting yesterday's decision, mixing private notes with public output, or citing the wrong source. Remanentia keeps the memory layer separate: documents, logs, facts, and vector evidence are indexed first; retrieval selects the relevant sources; Director Class AI then checks whether the answer is supported before it is used.
Verification before output
Retrieved evidence is treated as a constraint. The answer has to match the source material and the public/private policy, or the system should refuse, rewrite, or ask for more evidence.
Systems that remember their work
The target is not a prettier chat window. It is agent memory, customer knowledge bases, research assistants, operational handovers, and local AI infrastructure that can be audited.
Talk to the builder
Anulum CH&LI — Miroslav Šotek
Obergasse 8
9437 Marbach SG
Switzerland
Phone: +41 76 607 9990
Primary: protoscience@anulum.li
What to send first
The best conversations start with the failure mode. For example: the agent invents facts, retrieval returns stale policy, support answers leak internal notes, or local hardware cannot keep up.
- Whether the corpus is public, private, or mixed
- How often the source material changes
- Latency and accuracy limits that matter in practice
- Which agents or applications need memory
- Which information must never become public
Memory and verification deployments
Use this for product integrations, local deployments, corpus design, factual-control work, or a failing RAG system that needs a stricter memory layer.
protoscience@anulum.liResearch and benchmark scrutiny
Use this for formalism review, experiment design, benchmark methodology, paper feedback, or comparing Remanentia against other memory systems.
review@anulum.liInfrastructure and partnership
Use this for hardware, hosting, research sponsorship, strategic deployment support, or investment discussions tied to evidence-grounded AI systems.
invest@anulum.liPublic profiles
Support the work
Keep the memory, benchmark, and deployment work moving.
Remanentia needs the unglamorous parts too: servers, storage, benchmark runs, review time, and enough quiet engineering hours to make high-stakes LLM use less fragile. These channels are for direct support and infrastructure funding. Product licensing stays on the pricing page.
Direct infrastructure support
IBAN CHF: CH14 8080 8002 1898 7544 1
IBAN EUR: CH66 8080 8002 8173 6061 8
BIC: RAIFCH22
Reference: ANULUM Support
Long-horizon research support
BTC: bc1qg48gdmrjrjumn6fqltvt0cf0w6nvs0wggy37zd
ETH: 0xd9b07F617bEff4aC9CAdC2a13Dd631B1980905FF
LTC: ltc1q886tmvtlnj86kmg2urd8f5td3lmfh32xtpdrut
Small practical contributions
Use these when you want to help with compute, storage, benchmark runs, or public documentation without starting a commercial licensing conversation.