AnimeAtlas

An open anime identity and metadata database for offline lookup.

AnimeAtlas resolves anime names and external provider IDs to a stable series -> media -> episode identity chain. Media records carry verified Season/Part context, while episode numbers retain provider namespaces such as bangumi:ep and bangumi:sort.

Download the database

A single-file SQLite snapshot is published under a fixed download link that always points at the latest build — the URL never changes between releases:

https://github.com/ModerRAS/AnimeAtlas/releases/download/download/animeatlas.sqlite

curl -L -o animeatlas.sqlite \
  https://github.com/ModerRAS/AnimeAtlas/releases/download/download/animeatlas.sqlite

The download release remains a stable alias. Every publish also creates an immutable semantic-version release with animeatlas-<version>.sqlite and a SHA-256 manifest. Exact catalog, SQLite schema, generator, normalization, and parser revisions are embedded in release_info:

SELECT key, value FROM release_info ORDER BY key;

SQLite schema

series                 (id, title, relationships_json)
media                  (id, series_id, season_number, part_number, cour_number, metadata_json, provenance_json)
series_aliases         (series_id, value, normalized, language, type, source, confidence)
aliases                (media_id, value, normalized, language, type, source, confidence)
provider_refs          (media_id, provider, entity, provider_id, provider_key)
episodes               (id, media_id, kind, provenance_json)
episode_provider_refs  (episode_id, provider, entity, provider_id)
episode_numbers        (episode_id, namespace, number_value, source)
search_tokens           (token, media_id)
release_info            (key, value)

Aliases form candidate sets; duplicate normalized aliases across installments are valid. aliases_v1_compat exposes only aliases that resolve to one media. Episodes belong to one media, which guarantees that namespaced numbers cannot silently cross Seasons.

Query examples

-- Resolve an alias. `normalized` stores NFKC + trimmed + lowercased text.
SELECT m.id, m.title
FROM aliases a JOIN media m ON m.id = a.media_id
WHERE a.normalized = 'sousou no frieren';

-- Look up a media identity by a Bangumi subject ID.
SELECT m.id, m.title, m.summary
FROM provider_refs p JOIN media m ON m.id = p.media_id
WHERE p.provider = 'bangumi' AND p.entity = 'subject' AND p.provider_id = '400602';

-- Pull normalized metadata and its provenance for one media identity.
SELECT metadata_json, provenance_json FROM media WHERE id = 'media-000001';

Use the CLI

The repository also ships a CLI for offline resolution against the committed JSON indexes (no network needed).

Requirements: Node.js 22+, pnpm 10+.

corepack enable
pnpm install --frozen-lockfile
pnpm check

Resolve a title/alias or an external provider ID from the local indexes:

pnpm cli -- resolve alias "Tensei Shitara Slime Datta Ken" --season 4
pnpm cli -- resolve provider bangumi subject 515594

Both return a typed resolved, ambiguous, or unresolved result. Alias resolution reads the candidate-set index and accepts --season, --part, and --cour. Add --compact for single-line JSON output.

Command	Purpose
resolve alias <title> [--season N]	Resolve a candidate set with installment context
resolve provider <provider> <entity> <id>	Map an external provider ID to a media identity
bangumi plan-archive <file>	Plan a bulk import from a Bangumi archive dump
contributions plan-approved	Preview mutations from approved contribution Issues
contributions apply-approved --write	Apply approved contributions to db/

What's in the database

The committed snapshot is a seed dataset that consolidates IDs across providers and carries normalized metadata with field-level provenance (which provider, which source field, and which rule produced each value). It grows through reviewed community contributions. See generated/stats/summary.json for current record counts (media, aliases, provider refs, search tokens).

Data Model

reviewed source inputs
        |
        v
source/  ->  db/  ->  generated/  ->  SQLite release
                  ^
           normalized records

Directory	Purpose	Edit policy
source/	Approved community contributions, import manifests, and durable editorial decisions	Created through reviewed workflows
raw/	Optional captured provider evidence	Machine-written only
db/	Normalized media, alias, metadata, relation, and provenance records	Generated by the import pipeline
generated/	Deterministic lookup indexes, manifests, and statistics	Run pnpm generate; never edit manually
apps/	CLI, GitHub Action helper, and static viewer	Application entry points
packages/	Schemas, provider contracts, importer, validator, and generator	Reusable domain logic

generated/ is disposable output. source/ and provider evidence explain how the published snapshot was produced; db/ is the stable JSON consumption layer.

Contribute Data

Do not edit database JSON directly. Use the recognition or Season/episode Issue form. Contributors enter natural observations and a Bangumi target; automation validates Season ownership and converts optional text such as 06(78) into typed episode-number changes before approval.

1. A maintainer reviews the structured Issue and applies the approved label.
2. GitHub Actions parses the contribution, applies it through the importer, regenerates indexes, and runs pnpm check.
3. On success, automation commits the updated source/, db/, and generated/ records to master, closes the Issue, and refreshes the download SQLite release.

The approval label is the write gate. Community input is stored as an auditable contribution record before it affects normalized data.

Development

Command	Purpose
pnpm check	Build, typecheck, validate data, verify generated artifacts, and run smoke checks
pnpm validate	Validate source and normalized records
pnpm generate	Rebuild deterministic indexes and manifests from db/
pnpm check:generated	Fail when committed generated artifacts are stale
pnpm cli -- contributions plan-approved	Preview approved contribution mutations without writing files
pnpm cli -- contributions apply-approved --write	Apply approved contributions locally
pnpm release:sqlite	Build stable and immutable SQLite artifacts plus SHA-256 manifests
pnpm migrate:v2	Audit all curated db/migrations/v2-*.json plans; add -- --refresh for live evidence or -- --write to apply cached evidence
pnpm audit:v2	Cache and classify any remaining v1 Bangumi subjects, relations, and paginated regular episodes
pnpm replay:library	Compare v1/v2 against paths stored in read-only library.db

Run pnpm check before committing a data or schema change. It is the same validation gate used by repository automation.

Architecture

• AnimeAtlas v2 catalog
• Architecture overview
• Repository boundaries
• Schema-first design
• Validation and index generation
• GitHub automation

License

MIT