What it is
A public, synthetic proof of concept for a vault where small typed JSONL records are the canonical layer and Markdown or SQLite outputs are generated views.
Project page
A synthetic JSONL vault for agent-readable context.
A public proof of concept for replacing Markdown-only vault state with typed JSONL records, generated views, and bounded context bundles that agents can consume without guessing.
jsonl agent-context synthetic-data vault-schema sqlite
JSONL
canonical shape
Synthetic
data boundary
Agents
consumer
Project notes
Why it exists
Markdown is useful for humans, but agents need explicit structure, provenance, and bounded context. This repo tests that stricter record-first model.
Useful for
Agent context bundles, typed memory experiments, schema coverage tests, generated Markdown views, and local SQLite read models.
How it works
Synthetic raw evidence becomes typed records, retrieval hints, generated bundles, and browsable views. The examples stay public-safe by design.
Safety model
The repository intentionally avoids real personal vault exports, names, emails, paths, screenshots, credentials, or real attachments.
Current focus
Proving the record model, media and file references, generated views, validation gates, and 10,000-record synthetic dataset at repo scale.
Why it matters
It explores a cleaner interface between personal knowledge systems and AI agents: records first, views second, context always bounded.
Next action
Open GitHub when you want the current schemas, generators, tests, fixtures, and implementation details.
This page is the human-readable project entrypoint. GitHub remains the place for current files, implementation details, issues, and release state.