ingest

axiom-wiki ingest [file-or-url]
axiom-wiki ingest [file-or-url] --interactive

Ingest a local file, a URL, or scan raw/ for anything not yet processed.

Usage

axiom-wiki ingest path/to/file.pdf           # ingest a local file
axiom-wiki ingest https://example.com/article # clip URL and ingest in one step
axiom-wiki ingest                             # scan raw/ and ingest new files

When given a URL, Axiom fetches the page, extracts the article content (using Readability), saves it to raw/, and immediately ingests it — no separate clip step needed.

What happens

The agent reads the file, extracts entities and concepts, creates wiki pages, and updates the index. The terminal shows live progress:

write_page({"pagePath":"wiki/pages/entities/alan-turing.md"...})
  wiki/pages/entities/alan-turing.md written

my-notes.pdf
  in=42318 out=1847  $0.0231

+ wiki/pages/entities/alan-turing.md
+ wiki/pages/concepts/turing-test.md
~ wiki/index.md

Incremental compilation

Each source file is tracked by SHA-256 hash in state.json. When you run axiom-wiki ingest without a file argument, only new or modified sources are processed — unchanged files are skipped automatically. This makes re-running ingest fast even on large wikis.

Compilation lock

A PID-based lock prevents concurrent ingest operations. If another ingest is already running, you’ll see a “Compilation locked” message. Stale locks from crashed processes are automatically reclaimed.

Re-ingest

When you ingest a file that already has a wiki summary, Axiom compares old and new content and only updates pages that changed. You can also force a re-ingest from the sources screen by pressing r.

Interactive mode

Add --interactive to review topics before writing:

axiom-wiki ingest notes.md --interactive

See the Interactive Ingest guide for details.

Supported file types

.md, .txt, .pdf, .png, .jpg, .jpeg, .webp, .html, .docx