Research
Clio began as a question about memory, not as a note-taking product.
The system started from a simple concern: most AI memory stores information, but does very little to preserve the structure that makes information meaningful and usable. That led to a broader investigation into memory, meaning, and understanding in AI systems.
Clio is the product form of that investigation: a local system for building structured memory from your own material.
Core Thesis
Useful memory should preserve explanatory structure, not just retrievable fragments.
If memory loses the relationships that make material intelligible, it becomes weaker exactly when the question gets harder.
Meaning depends on how pieces of information explain and constrain one another.
A memory system is weaker when everything is reduced to isolated passages.
For harder questions, retrieval must preserve more than textual proximity.
Limits of Retrieval
Finding similar text is not the same as recovering structure.
Similarity-based retrieval works best when the right passage already exists in a form that closely matches the query. It works less well when the answer is distributed across sources, expressed differently, or dependent on relationships not visible at the surface level.
The relevant material may be spread across many documents, notes, or prior situations.
The right material may use different words than the query even when it shares the same structure.
A flat store can return passages, but it does not preserve which ideas organize which facts.
The Memory Model
Built around concepts, facts, and linked situations — and an agent that constructs them.
Clio takes one practical step: build a local memory system that preserves more
structure than standard retrieval layers do. The memory model is the theory.
Clio Agent and .clio/ are what it becomes in practice.
Organizing structures that make separate pieces of evidence belong to the same intelligible pattern.
Concrete observations and details that become more useful when attached to the concepts that contextualize them.
Prior cases and structurally similar moments that can be surfaced when a new problem shares their shape.
Performs the mapping work. Reads source material, extracts concepts and facts, and writes structure into local memory using local or hosted model backends.
The durable memory system that remains: the local files inside .clio/ that can be queried through CLI and MCP by users and assistants.