Skip to navigation
File SearchFile Types

Supported File Types

Captain indexes the 186 file types below, sent through a direct upload, the dashboard, raw text, or a cloud storage connector. Each one is parsed, chunked, and embedded for search.

CategoryCountExtensions
Documents3pdf, docx, doc
Sheets4xlsx, xls, csv, tsv
Images7png, jpg, jpeg, gif, bmp, tiff, webp
Audio7mp3, wav, m4a, ogg, flac, aac, wma
Video7mp4, mov, avi, mkv, webm, flv, wmv
Email2msg, eml
Text and code156txt, md, json, yaml, and 151 source, markup, and config extensions listed under Text and code

Documents

How the document lane works, with citations and the two processing tiers: Documents.

ExtensionTypeWhat you get
.pdfPDFText chunks labeled by layout role (body, table, heading, figure, and more), each with a page range. See Advanced Search & Relations.
.docx, .docWordText chunks with a page range (metadata.pageStart/pageEnd).

Sheets

How workbooks become row bands with provenance: Sheets.

ExtensionTypeWhat you get
.xlsx, .xlsExcelStructure-aware chunks with row/column provenance (sheetName, rowStart/rowEnd, colStart/colEnd, columns). See Multimodal Search.
.csv, .tsvTabular textSame row/column provenance as Excel.

Text and code

Chunking, JSON parsing scripts and the raw text route: Text and Code.

Every extension in this section is read as plain text and split into text chunks with a page range. There is no syntax-aware parsing: a .py file and a .txt file go through the same lane. For .json files, pass a parsing_script to control chunking.

FamilyExtensions
Plain text and data.txt, .md, .json, .yaml, .yml
Python.py, .pyi, .pyw, .ipynb
JavaScript and TypeScript.js, .mjs, .cjs, .jsx, .ts, .tsx, .mts, .cts, .coffee
JVM.java, .kt, .kts, .scala, .sbt, .groovy, .gradle, .clj, .cljs, .edn
C family.c, .h, .cpp, .cc, .cxx, .hpp, .hxx, .hh, .m, .mm
.NET.cs, .csx, .fs, .fsx, .razor, .cshtml
Other languages.go, .rs, .rb, .erb, .rake, .gemspec, .php, .phtml, .swift, .pl, .pm, .lua, .r, .rmd, .dart, .jl, .nim, .zig, .ex, .exs, .erl, .hrl, .hs, .elm, .ml, .mli, .s, .asm, .vbs, .applescript
SQL.sql, .ddl
Shell and scripts.sh, .bash, .zsh, .fish, .ksh, .csh, .tcsh, .ps1, .psm1, .psd1, .bat, .cmd
Web, markup, and templates.html, .htm, .xhtml, .shtml, .css, .scss, .sass, .less, .styl, .vue, .svelte, .astro, .hbs, .ejs, .pug, .twig, .liquid, .njk, .mustache
Markdown and docs.markdown, .mdown, .mdx, .rst, .adoc, .asciidoc, .org, .tex, .bib, .textile, .text, .diff, .patch
Config and infra.toml, .ini, .cfg, .conf, .properties, .lock, .tf, .tfvars, .hcl, .nix, .cmake, .mk, .bzl, .bazel, .dockerfile, .log
JSON variants and data text.json5, .jsonc, .jsonl, .ndjson, .geojson, .graphql, .gql, .proto
XML family.xml, .xsl, .xslt, .xsd, .dtd, .rss, .atom, .xaml, .svg, .drawio, .csproj, .sln

Not indexed, by design: compiled or binary variants (.pyc, .pyo, .class, .o, .so, .dll, .wasm, .jar, .svgz), .plist (often binary), and secret-bearing files (.env, .pem, .key, .crt).

Files with no extension arriving through a cloud storage connector are not rejected up front. The indexer sniffs their content type (HTTP content type plus magic bytes) and indexes them when the content is a supported type.

Images

What an image becomes and how mask_pii treats it: Images.

ExtensionTypeWhat you get
.png, .jpg, .jpeg, .gif, .bmp, .tiff, .webpRaster imagesOne chunk per image: a native embedding of the pixels plus a written description stored as searchable text, so a query matches what the image shows, not just its filename.

Images are supported up to 200 megapixels decoded, and can run through mask_pii for OCR-based redaction. See PII Masking.

Video

Clips, transcripts and descriptions: Video.

ExtensionTypeWhat you get
.mp4, .mov, .avi, .mkv, .webm, .flv, .wmvVideoSegmented into clips up to 120s. Each clip has a native embedding plus a transcript and a visual description stored as searchable text. Segment time range in metadata.startSec/endSec.

Audio

Clips and transcripts for audio files: Audio.

ExtensionTypeWhat you get
.mp3, .wav, .aac, .flac, .m4a, .ogg, .wmaAudioSegmented into clips up to 80s. Each clip has a native embedding plus a transcript and a description stored as searchable text. Segment time range in metadata.startSec/endSec.

Video and audio detail, including transcript language control, lives on Multimodal Search.

Email

Parts, relations and metadata in full: Email.

ExtensionTypeWhat you get
.msg, .emlOutlook / RFC 822 messageThe message body as a text document plus one document per attachment, each parsed as its own type. email_attachment relations link the body to its attachments and email_reply_to links a reply to its parent, in both directions. See Advanced Search & Relations.

Every part carries these filterable metadata keys (empty string when the header is absent):

KeyOnValue
email_roleallbody or attachment
email_message_id, email_in_reply_toall / bodyMessage-ID without angle brackets
email_thread_idallMessage-ID at the root of the reply chain, shared by the whole thread
email_from, email_subject, email_dateallHeaders as sent; date as ISO 8601 UTC
email_source_uriallThe email file’s own URI (s3://…, gs://…, blob URL, dropbox://…)
email_parent_identityattachmentsThe body’s identity, msg://<Message-ID>
email_attachment_name, email_content_sha256attachmentsFile name; SHA-256 of the bytes
email_attachment_countbodyAttachments indexed from the message

Not indexed: inline signature images, empty or duplicate attachments, and unsupported types. Re-indexing an email replaces its parts in place and removes attachments no longer present.

{ "query": "unit price", "filter": { "email_thread_id": { "$eq": "quote-2026-09-01@example.com" } } }

Size limits by type

Each type has its own per-file and per-request ceiling, and the ceiling also depends on how the file reaches Captain (dashboard, API, or a connector). See File Size Limits for the full breakdown.