OUTSIDE.

The starting point was a question about a system that preserves its own history and develops through experience. I designed separate layers for the organism, persistent storage, information providers, a supervisor, and an observer. Starting the program and creating a new identity are different operations. This allows a later code version to develop the existing system without pretending that its earlier life has begun again.

I built SQLite storage, data-structure versioning, backups, and a history of changes. Birth identity remains separate from the current software version. The system records the passage of time and confirmations of its activity; after an interruption, it preserves uncertainty instead of adding experiences from a period it did not observe. I also separated real-world state from experimental environments so that a trial of a new mechanism does not alter its history.

I developed a layer that connects observations with predictions and compares an expected outcome with new experience. What matters is not only the content of information, but also its source, freshness, and reliability. A contradiction creates a reason to review a belief. A language model’s statement retains its own provenance and does not automatically become evidence about the world. This separation makes it possible to study how the system knows something and when it should change its mind.

I developed mechanisms for homeostasis, deriving needs, selecting actions, and habituation. The system’s state, available capabilities, and previous outcomes affect the assessment of the next step. A repetitive signal can lose importance for attention, while new information can regain it. The work also includes a model of its own capabilities and a decision trail. This makes it possible to determine what problem the system was trying to solve and whether it had the right tool.

I prepared an observer that presents persisted events and research state through a read-only interface. The view is intended to explain the system’s process, not control its behavior. I also separated the recording of facts from interpretation. This allows me to return to sources, see change over time, and study mechanisms without replacing actual experiences with a compelling story.

Development includes modules for research questions, evidence assessment, experience review, and designing possible improvements. For new capabilities, I prepared contracts, persistent records, and test scenarios. I verify successive stages in an isolated environment. The mere presence of a module does not mean it has been activated in the operating world; the research includes both completed mechanisms and experiments being prepared.

OUTSIDE is my own project and the source of algorithms I developed. In Reachy, I develop selected mechanisms for memory, attention, homeostasis, evidence assessment, and learning, combining them with conversation and relationships. These projects have separate contexts and histories. Their shared body of work makes it possible to carry over verified solutions and ask new questions without reducing the entire effort to a single conversational interface.