Domain Foundry
Turn an interest into an app you own, with its evidence attached
shippedagent2026
Platform
Local Mac app: Python core, FastAPI, React shell, SQLite. CLI, MCP, Telegram, hermes-agent as front doors. github.com/finnqiao/domain_foundry
What it does?
Capture notes into structured domain records on your machine, and compile a real practice into an app you own, with the evidence attached.
Highlights
- ▸Type a note (or ingest a folder / Obsidian vault). Rules file the easy ones. A model only runs when the capture is ambiguous, and a failed model still keeps the note.
- ▸Describe a practice you care about. Compare three structurally different product cuts, then compile the one you pick into a local app whose tables come from named workloads.
- ▸Point import at an existing SQLite database. Dry-run shows where every row lands. Apply never writes the source (opened read-only). This is how a real Hermes dataset moved over.
Architecture
One HarnessAPI on localhost. Two SQLite DBs: append-only ledger.sqlite and typed domains.sqlite. L1 regex then L2 LLM. FoundrySpec compiles research, schema, preview, owned app, and a hashed build receipt from one contract. Bring-your-own key, two model tiers, no telemetry. Release candidate, not a published package.
Stack
PythonSQLiteFastAPIReactMCP