Solaris v0.1.0-alpha — An Archive-Grounded Memory Substrate

Solaris is an archive-grounded editorial memory substrate for long-lived AI systems.

It archives broadly, derives structure cautiously, remembers selectively through editorial judgment, and retrieves with provenance. The design goal is memory that stays inspectable and honest instead of collapsing everything into fuzzy recall.

This first public release includes:

• Append-only event archive with inspectable provenance
• Editorial review over derived memory
• Archive-vs-belief divergence
• Graph-aware retrieval
• Computed pattern projection
• Host shadow-mode examples
• Backfill tooling for generic host exports
• Committed eval and proof runners

Repository highlights:

• run_retrieval_evals.py for retrieval regression checks
• run_archive_state_model_checks.py for archive-state proof checks
• docs/ for architecture, usage, proof, and integration notes
• examples/ and tools/ for sidecar/shadow-mode adoption

Status:

• alpha / research preview
• intended for builders and researchers
• not positioned as a polished end-user product yet

License:

• MPL-2.0