The memory layer for AI.

A system built to resolve contradictions at scale.

Join the beta Give feedback Read the documentation

Standard MCP connection

Compatible with any client.

Your data belongs to you. Full control.

Chat. It remembers everything. Even what you forgot.: Your contractor changes address. Then phone number. Then address again. Six months and 2,000 messages later, someone asks for the current address. Only one fact counts: the last one.; What this means : When something changes, AMAFOS retrieves the version that holds today, and keeps the full history: nothing is ever overwritten.
Import. A lease, a report, a spec doc. 6 months later, it finds the clause.: An 80-page lease imported in June. A 200-page technical spec in July. In November, someone asks for the termination clause or the load tolerance. The right passage has to surface, not a summary.; What this means : Across hundreds of thousands of tokens of raw text, AMAFOS finds the exact passage, not a summary that looks like it. The search covers the whole document, not just what still fits in the context window.
Track. Budget goes from 3,500 to 4,200 €. It sees it. It tells you.: As the conversation goes, you set rules: “invoices above 5,000 go to director approval”. The system has to hold on to the rule and apply it to later cases, even weeks afterwards.; What this means : AMAFOS learns your business rules as the conversation goes, keeps them beyond the context window, and applies them — weeks later included.
Reason. Two updated facts, one question. It chains both.: The project architect changes. The schedule depends on the architect. Giving the new delivery date means chaining both updates.; What this means : The most demanding test in the benchmark. Retrieving a fact is not enough: it takes reasoning across several changes tied to one another. AMAFOS chains them.
Understand. Your character is left-handed in chapter 2. Right-handed in chapter 7. It catches that.: Your character is left-handed in chapter 2. Right-handed in chapter 7. The inconsistency has to be caught, not just the word “left-handed” retrieved. The system has to grasp meaning, not match strings.; What this means : AMAFOS cross-references keyword search (BM25) and semantic search (embeddings) to hold a global understanding across very long contexts. It is the combination of the two that makes the difference.