The Ingestor

The Ingestor is the agent that runs every time you upload a file to BaseForge. It reads the file, breaks it into searchable pieces, and writes a summary back to the file record. Until the Ingestor finishes, other agents (the Writer, the Strategist, the Mapper) can't use that file as a source.

What it does to each file

The Ingestor takes a file you've uploaded and produces:

• A document summary. A short paragraph describing what the file is about.
• A list of key topics. Up to seven topical tags pulled from the content.
• A document type. A best-guess label (e.g. case study, product page, whitepaper).
• Knowledge chunks. The file is split into roughly paragraph-sized pieces. Each chunk carries the section it came from, the entities mentioned in it, and topical tags.
• Embeddings. Each chunk gets a vector representation so other agents can pull it up by semantic similarity when they need to ground a draft or build the entity map.

You don't interact with these directly — they're what makes the rest of the platform work on your content.

When it runs

The Ingestor runs automatically:

• When you upload a file through the BaseForge UI.

It also re-runs if you re-process a file. Re-processing is safe — the Ingestor deletes the old chunks for that file before writing new ones, so you won't end up with duplicates.

What you'll see in BaseForge

While the Ingestor is working, the file shows as processing in your document list. When it finishes:

• The file flips to ready.
• The summary appears on the file's detail view.
• The chunk count is recorded on the file.
• The file becomes available as a source for new content drafts and is included in the next entity map rebuild.

If a file fails to ingest

If the Ingestor can't process a file, its status moves to error and an error message is attached to the file record. The most common reasons:

• The file has no extractable text (e.g. a scanned PDF with no OCR layer, an image-only document).
• The file is empty or corrupted.
• The connection to the AI provider failed mid-process.

Customising the prompts

Admins can override the prompts the Ingestor uses for structuring, metadata extraction, and summary generation under Settings → Agent Prompts → Ingestor. The three editable prompts are:

• Structuring — how the Ingestor decides where sections begin and end in a long document.
• Metadata extraction — how it picks the entities and topics that get attached to each chunk.
• Summary — how it writes the file-level summary you see in BaseForge.

An admin can save a brand-specific override and reset back to the platform default at any time.