Household Knowledge Base (Synthetic Documents)

A collection of synthetic but realistic household-property documents — service receipts, contractor emails, manuals, inspection reports, utility bills, mortgage statements, HOA notices, neighborhood communications — formatted for demonstration of personal-property annotation and decisional-aggregate workflows with Semiont.

This KB is the operational counterpart to the home-management worked examples in semiont-template-kb's PERSONAL-KB-EXPLORATION.md. The skills implement those queries — backward-looking maintenance assessments, forward-looking system-priority lists, vendor track-records — against synthetic data, but operate generically over any household-property corpus dropped into the same directory layout.

About This Dataset

This repository contains synthetic household-property materials. All addresses, vendors, contractor names, account numbers, dates, model numbers, and dollar amounts are entirely fictional. The documents resemble real homeowner records in form and shape but describe no actual property, no actual people, and no actual transactions.

The corpus covers the full surface area of a homeowner's documentation:

• Rooms — kitchen, baths, bedrooms, basement, attic, garage, yard
• Subsystems — HVAC, plumbing, electrical, roofing, foundation, networking, security, entertainment
• Utilities — electric, gas, water, internet, garbage / recycling
• Recurring services — pest control, lawn care, gutter cleaning, exterior washing, HVAC tune-ups, chimney sweeping, snow removal
• Episodic services — bat removal, roof repair, plumbing emergencies, appliance replacement
• Appliances — laundry (washer, dryer), kitchen, water heater, sump pump
• Mortgage and insurance — lender statements, escrow, refinancing notes, homeowner's policy, property-tax assessments
• Neighborhood / HOA — HOA notices, neighborhood-listserv excerpts, easement / right-of-way notes

Skills

This repo ships ten skills that build a layered home-property KB on top of the Semiont SDK. See AGENTS.md for the full design discussion.

Skill	What it does
ingest-corpus	Walk the repo's corpus and create one resource per file.
mark-house-entities	Detect Person, Room, Subsystem, Appliance, Vendor, Utility, Service, Date, MonetaryValue, Address spans — including anaphoric mentions ("the contractor", "the AC unit", "the master bath").
canonicalize-rooms	Promote Room mentions to canonical Room resources.
canonicalize-subsystems	Promote Subsystem / Appliance mentions to canonical Subsystem resources, with installed-date / warranty / manufacturer-rated lifespan.
canonicalize-vendors	Promote Vendor / service-company mentions to canonical Vendor resources.
extract-events	Synthesize one Event resource per dated service / repair / payment / visit, bound to its Subsystem and Vendor.
track-recurring-services	Identify recurring service schedules (quarterly pest control, biannual HVAC); project next-due dates.
comment-action-items	Surface follow-up items, missing documentation, unaddressed warranty events.
assess-house-maintenance	Backward-looking aggregate — per-Subsystem maintenance score, deferred work, documentation gaps.
prioritize-house-systems	Forward-looking aggregate — ranked priority list with cost estimate, cascade risk, warranty-status.
vendor-track-record	Per-Vendor decisional summary — visit history, reliability, costs, last-used recency.

Quick Start

Explore this dataset using Semiont. This repo follows the same layout and startup flow as semiont-template-kb.

Open in Codespaces

Prerequisites: the Semiont launcher (brew install the-ai-alliance/semiont/semiont) and the GitHub CLI (gh), signed in with gh auth login.

Before creating: add ANTHROPIC_API_KEY as a user secret with this repo selected. Otherwise the backend comes up but inference is non-functional until you add the secret and rebuild the container.

One command creates the codespace (or resumes the one you already have), waits for the stack to answer, and forwards the KB to your machine:

semiont start --runtime codespace --repo The-AI-Alliance/semiont-household-kb

No account exists until you make one — the same as a local stack (it prompts for the password):

semiont useradd --repo The-AI-Alliance/semiont-household-kb --email you@example.com --admin

The browser runs locally and connects to any number of knowledge bases — cloud or local:

semiont start --service frontend

Open http://localhost:3000 and add the KB in the Knowledge Bases panel, using the port the launcher printed and the credentials you just created. semiont stop --repo The-AI-Alliance/semiont-household-kb halts billing and keeps your state; add --delete to destroy the codespace.

Without the launcher: the raw gh recipe

gh codespace create --repo The-AI-Alliance/semiont-household-kb --machine premiumLinux
gh codespace ports forward 3000:3000 4000:4000   # leave running

# In another terminal, create the first admin (nothing creates one for you).
# --generate-password prints a random password once; there is no --password flag:
gh codespace ssh -- 'cd /workspaces/* && docker compose -f .semiont/compose/backend.yml \
  exec -T backend semiont-useradd --email you@example.com --generate-password --admin'

This forwards the codespace's own browser as well, so you open http://localhost:3000 and sign in as the admin you just created. If gh rejects the forward with must have admin rights to Repository, grant the scope once: gh auth refresh -h github.com -s codespace.

Privacy. This synthetic dataset poses no privacy risk; a real personal-property KB deployment with actual addresses, vendor contracts, and mortgage details should use the local-first start path rather than Codespaces. See PERSONAL-KB-EXPLORATION.md for the longer discussion.

License

Apache 2.0 — See LICENSE for details.