Look and Feel

Attachments

Attach files directly to your chat messages. Sparky handles each file type intelligently — images are sent natively to the LLM, while documents and code are converted to markdown and included as inline text.

How It Works

Images

Image files (PNG, JPG, JPEG, GIF, WebP) are sent directly to the LLM as visual content. The model sees the actual image and can describe, analyze, or extract information from it.

Documents & Code

All other supported files — PDFs, Word documents, spreadsheets, code, config files, and more — are converted to markdown using markit before being sent to the LLM. This means the model receives clean, structured text it can reason about.

For example:

• A PDF is extracted to text with headings preserved
• A DOCX keeps its heading structure and tables
• An XLSX becomes markdown tables (one per sheet)
• A CSV becomes a markdown table
• Code files are wrapped in fenced code blocks with language detection
• JSON/YAML are formatted as code blocks

Size Limit

Documents that exceed the configured character limit (default: 100K characters, ~25K tokens) are rejected with a helpful message suggesting you add the file to the Knowledge Base instead. This prevents a single large document from consuming the entire context window.

You can adjust this limit in Settings → Converters → Max output size (range: 10K–200K characters).

Adding Attachments

There are two ways to attach files:

1. Click the attachment button (📎) in the chat input toolbar — opens a file picker filtered to supported formats
2. Drag and drop — drag files from your file manager directly onto the Sparky window

Attached files appear as thumbnails above the input area. Click the ✕ to remove an attachment before sending.

Supported Formats

Category	Formats
Images	PNG, JPG, JPEG, GIF, WebP
Documents	PDF, DOCX, PPTX, XLSX, EPUB
Data	CSV, TSV, JSON, YAML, XML
Web	HTML, HTM, RSS, Atom
Code	Python, TypeScript, JavaScript, Rust, Go, Java, C/C++, C#, Swift, Kotlin, Ruby, PHP, SQL, and many more
Text	TXT, Markdown, RST, LOG, TOML, INI, ENV
Archives	ZIP (contents extracted recursively)
Notebooks	Jupyter (.ipynb)

Large Documents

For documents too large to include inline (books, lengthy reports, codebases), use the Knowledge Base instead. The Knowledge Base:

• Chunks the document into smaller sections
• Indexes them with BM25 full-text search and optional vector embeddings
• Retrieves only the relevant sections for each query

This is far more efficient than sending an entire document in every message.

Converter Settings

Adjust attachment behavior in Settings → Converters:

Setting	Default	Description
Max output size	100K chars	Maximum converted document size for chat attachments

URL extractor settings (for the Knowledge Base) are also available on this page — see Knowledge Base for details.