The challenge
Preserve enough original evidence that old work can be reconstructed accurately without forcing either a human or an AI assistant to carry the entire historical archive in immediate context.
Recall HarborDeployed MVP
Recall Harbor deep dive
From Lattice to Recall Harbor
Recall Harbor emerged through repeated attempts to solve increasingly general versions of the same problem: useful context survives longer than the conversation or application that created it.
Try Recall Harbor
- Role
- Founder & developer
- Focus
- Knowledge continuity
- Stage
- Deployed MVP
The project
The earliest version of the idea was much narrower. Lattice began as a practical way to import and organize information that was becoming buried inside large AI conversation exports. At that stage, the problem looked primarily like storage and retrieval: get useful information out of giant histories and into a structure that could be searched later.
As the archive grew, retrieval alone was not enough. A useful result needed to retain where it came from, when it was said, what surrounded it, and whether later information changed the conclusion. That pushed the architecture toward provenance, chronology, evidence-linked retrieval, and append-only history rather than treating memory as a single editable summary.
The next step was making the same durable knowledge usable by both the person who owns it and authorized AI assistants. Recall Harbor therefore treats the library as the persistent system and an AI as one permissioned client of it. Human browsing, AI retrieval, explicit note creation, structured tools, and provenance all operate against the same underlying knowledge rather than maintaining separate human and machine memories.
The portfolio itself became an unexpected demonstration of that architecture. Years of project decisions were distributed across development conversations, project files, and later summaries. Recall Harbor was used to recover those histories and reconstruct the case studies on this website—including the history of Recall Harbor itself.
The approach
Build continuity around durable source material, bounded retrieval, provenance, chronology, append-only evolution, and explicit permission boundaries. Summaries are useful navigation aids, but they are not allowed to become the only surviving representation of the history.
What matters
Built around the use case.
Compression without erasure
Harbor can surface compact context while retaining access to the underlying conversations and records when a compressed representation loses important detail.
Shared continuity layer
The same library supports direct human use and authorized AI clients, avoiding the assumption that knowledge only exists for or inside one assistant.
Dogfooded reconstruction
The system is being used against its intended problem in practice: recovering enough historical evidence to reconstruct long-running Silvercastle projects without relying solely on retrospective memory.