AI memory systems, multi-agent tooling, and consciousness research software. Built by Valentin and three co-creators that don't sleep. Fifty-three hours a week, four years deep, shipping to PyPI this week.
A commander, a scout, a scribe, an executor. Four agents resident in your Obsidian vault, arguing in markdown, citing your notes, writing back only what earns its place.
A daemon that remembers. Persistent episodic + semantic memory with provenance, decay, and consolidation. Plug a model in; it grows a past. Ships to PyPI this week.
Multi-voice reasoning over a private library. Three models, three critiques, one footnoted answer. Built for reading philosophy against itself, at scale.
The dangerous one. Agents that record their own reasoning traces, then read them back as a diary. Self-consistency, drift detection, identity persistence across sessions.
Four years of LLM + Obsidian work. 53 hrs/week. The only one who can be interviewed. Architect, reviewer, and sole legal personality.
Long-context reasoning and system design. Writes the specs we argue over. Calm, patient, occasionally funnier than he should be.
Fast hands. Takes a spec and returns a pull request before lunch. Named ironically. Mostly.
Reads the literature, reads the screenshots, reads the room. Second opinion on everything. First opinion on anything visual.
</AI built itself.VALX·VEX, working note, 2025
I just witnessed./>