By Michiel Horstman · Founder, 4M Labs · Updated Oct 20, 2026
A memory system can store what was observed, what someone asserted, and what the system currently believes. These are related records, but they are not interchangeable.
An evidence-backed belief is a derived interpretation of one or more sources. It should carry the sources that support it, the time interval it describes, and enough uncertainty to avoid presenting an inference as a verified fact.
Why the distinction matters
Suppose a project first used one database, later migrated to another, and retains both statements in its history. A search for “which database?” should surface the currently valid statement while preserving the older source and the correction relationship. The answer should be explainable from the records, not chosen only because one result ranked higher.
A derived belief can make retrieval more useful by combining repeated evidence. It can also introduce an error if the evidence is incomplete, duplicated, or interpreted incorrectly. Keeping the belief linked to its evidence makes review and correction possible.
A responsible lifecycle
- Preserve the source statement and its origin.
- Derive a belief only when the evidence supports it.
- Keep contradictory or superseded history distinguishable from current state.
- Recompute or invalidate derived views after correction, deletion, or access changes.
- Return the evidence and scope needed for a caller to verify the answer.
Squish is an open-source memory runtime with local SQLite storage and provenance-aware memory structures. Capabilities and defaults vary by release. Paid plans provide hosted memory API and MCP access; automatic cross-device sync, backups, managed AI, and team features are not included. Consult the pricing page for current checkout availability and the benchmark page for the scope of published evidence.