kibisis.dev

kibisis.dev is an experimental semantic atlas for classical Greek and Roman literature. It combines hybrid search, metadata filters, 3D semantic visualization, and source-grounded generated answers so readers can move from an idea to authors, works, passages, and cited textual evidence.

Live site: kibisis.dev

The public corpus is derived from PerseusDL / Tufts University and related canonical Greek and Latin data. The app only publishes passages classified as CC-compatible, and keeps provenance visible in the interface and API responses.

Why Kibisis?

In the Perseus myth, the kibisis is the pouch or satchel given to the hero to carry Medusa's head. The name fits this portal as a careful container for powerful textual fragments: passages, sources, metadata, and generated interpretations stay bound to their provenance.

What It Does

• Search ideas, scenes, authors, and works across classical texts.
• Navigate the corpus hierarchically from authors to works to passages.
• Inspect semantic relationships in a 3D atlas.
• Generate answers constrained to selected passages and cited sources.
• Preserve source metadata such as author, work, passage reference, CTS URN, license status, and dataset snapshot.

How To Use The Site

Start at kibisis.dev and search for an idea, scene, author, or work. You can also begin by exploring the authors directly: authors are positioned semantically in relation to their works, and selecting an author lets you narrow to that author's works before reaching individual passages.

Every visible point is organized semantically. A point marks the tip of a vector representing the meaning of an author, work, or passage, depending on the current map level. Moving through the hierarchy keeps the exploration fast and interpretable: author to work to passage.

Use the metadata filters to refine by genre, period, language, text type, author, or work. Switch the map level when you want a broader author/work view or a passage-level view inside a smaller scope.

Open passages from the results or the 3D atlas to inspect source metadata and provenance. When you have selected useful passages, generate an answer to produce a source-grounded synthesis based only on that context.

Support

kibisis.dev has ongoing server, generative model, and embedding costs. If the project is useful to you, you can help keep it online through Buy me a coffee.

Follow project updates on X / Twitter.

Repository Layout

• app/ - Next.js application and API routes.
• pipeline/ - corpus ingestion, classification, embedding, and export pipeline.
• infra/ - local database and migration utilities.
• brand/ - project marks, lockups, and visual assets.

Local corpora, generated data, agent files, and environment files should stay outside the public repository.

Quickstart

Prerequisites: Docker, Node.js 22+, npm, Python 3.12, and uv.

1. Install app dependencies:

cd app
npm install

2. Start local Postgres with pgvector:

docker compose -f infra/docker-compose.yml up -d

3. Apply migrations:

DATABASE_URL=postgresql://perseus:perseus@localhost:5432/perseus infra/apply-migrations.sh

4. Run the app with the root .env exported:

cd app
set -a && source ../.env && set +a && npm run dev

Open http://localhost:3000/pt.

Pipeline

Place upstream corpora under data-sources/ before running the ingestion pipeline:

• PerseusDL/canonical-greekLit as data-sources/canonical-greekLit or data-sources/greekLit_data
• PerseusDL/canonical-latinLit as data-sources/canonical-latinLit or data-sources/latinLit_data
• PerseusDL/catalog as data-sources/catalog or data-sources/catalog_data

Then run a small local pipeline pass:

cd pipeline
uv run pytest
PYTHONPATH=src uv run python scripts/run_pipeline.py --language eng --cc-only --limit 10 --persist

Database Compatibility

Development defaults to local Postgres with pgvector and pg_trgm. Neon remains compatible because the app and pipeline depend on DATABASE_URL, standard Postgres SQL, full-text GIN indexes, and pgvector.

Vercel + Neon

When using the Neon integration in Vercel, set the Vercel project root directory to app.

Use the pooled DATABASE_URL for the deployed app. Use the direct DATABASE_URL_UNPOOLED for schema migrations, data restores, and pipeline writes.

After linking Neon to the Vercel project, pull the production environment locally:

cd app
vercel env pull .env.production.local --environment=production
set -a && source .env.production.local && set +a
cd ..

Apply schema and load a data dump into Neon:

infra/dump-data.sh
DATABASE_URL_UNPOOLED="$DATABASE_URL_UNPOOLED" infra/bootstrap-neon.sh

For a repeat load into an already populated target, opt in explicitly:

RESET_DATA=true DATABASE_URL_UNPOOLED="$DATABASE_URL_UNPOOLED" infra/load-data.sh

License Governance

The raw pipeline may ingest all corpus files for audit and diagnostics, but the API and static export publish only passages classified as cc_compatible. Unknown or restricted files are default-denied.

Generated answers are an exploration aid, not a critical edition or final scholarly judgment. Always verify cited passages against the listed sources.