# ontology-docs
A standalone reference datapack demonstrating Ontology's native executable,
JSON/MCP interface, original image and discoverable/exportable Agent Skills.
```sh
cargo install ontology-docs
ontology-docs
ontology-docs search 'password protection'
ontology-docs version
ontology-docs sanitization
ontology-docs provenance document:overview
ontology-docs tables
ontology-docs query --sql 'SELECT media_type,count(*) AS documents FROM documents GROUP BY media_type'
ontology-docs export --format parquet --table documents --output documents.parquet
ontology-docs export --format sqlite --output analytics.sqlite
ontology-docs images
ontology-docs logs list
ontology-docs logs search 'document graph'
ontology-docs skills list
ontology-docs skills read ontology-authoring
ontology-docs skills export ontology-authoring --output ./agent-skills
ontology-docs mcp-config
```
Normal commands return `{ok,command,data,next}` JSON; follow the actual next argv
arrays for cited reads, image fetches and skill export. `--human` opts into text.
`serve` is newline-delimited stdio MCP. No runtime files, network, compiler or
embedding service is needed. This is a target-specific native executable, with
normal platform runtime requirements. WASM is future work.
The embedded demo contains portable ontology:// provenance,
the root README, format/agent/HTTP/ACP packaging and Notes guides, an original 32x24 gold/teal PNG, and three
skills: ontology-skill-creator, ontology-authoring and ontology-docs-use. The grid
was created directly from alternating RGB blocks; it has no third-party image
source. Runtime image embeddings are deliberately absent; import aligned external
vectors when needed rather than claim a bundled diffusion encoder.
The demo also includes browsable source and bounded text excerpts from six frozen
development sessions. Screenshots are omitted from history; omission counts,
selection rules and source checksums are bundled. Building/installing never reads
live harness logs. Its single embedded data file has a 9 MiB budget; no separate
data crates or runtime downloads are required.
Regenerate from the repository root after updating docs or skills:
```sh
cargo run --example generate_docs --locked
cargo build -p ontology-docs --release --locked
```
The Rust generator uses library builders and the native skill linter. The binary
assembles five version-pinned data-only build dependencies and includes the resulting
fixture with include_bytes. These keep each Cargo upload below the registry limit;
the installed executable needs no data files or downloads. Generation runs only while
authoring the crate, never on installation or at runtime. Export writes the chosen
ROOT/skill-name and does not install skills or change harness configuration.
`ontology-docs server --port 0` launches the embedded dashboard and shared HTTP API.
The demo includes current competition/usage guides, original binary bytes and a
nested datapack; explore `objects`, `graph`, `embeddings` and `packages`.
Version and Provenance tabs expose packaging identity, full checksums, original document metadata, Git author/history records and source tracing. Compression research and local log benchmarks are bundled as searchable documentation.
The six Codex project log files, including the session that built Ontology, are
included as frozen, compressed JSONL snapshots, with searchable transcript previews and native full-record retrieval.
`read creation-session` explains the archive; `logs list` discovers its session ID.
`object asset:creation-session-manifest` identifies the capture cutoff, checksums
and recorded omissions. Private reasoning and internal harness instructions are
omitted. The archives are embedded in the executable and ship with this crate.
The source snapshot is included as a ZIP and individual files: use the human
Browse tab to walk folders, preview source and download retained copies.
The sanitization guide and shared default rules are embedded. The demo explicitly
keeps its public fixture images and binary examples as uninspected bytes; text,
source metadata and frozen log records are sanitized. Every replacement uses
`<redacted by ontology>`. Source manifests distinguish bundled hashes from source
hashes, and snapshot manifests record sanitization coverage. Zero replacements is
not a clean-corpus guarantee. Newly authored datapacks first create editable
`ontology-sanitize.toml`; review/edit it and repeat the packaging command.
Analytics/export and Obsidian vault guides are also embedded. Exports support
filtered tables or read-only SQL results through CLI, MCP and HTTP. Built-in
formats work offline; native DuckDB creation uses a local DuckDB CLI, while its
Parquet import bundle needs no external tool to export.
Datapack is Ontology's unit of distribution: data, provenance, indexes, skills and
runtime in one executable. The GitHub source guide is embedded; search GitHub to
learn how to package a repository or select a revision and folder.