Skip to content

Releases

v0.4.1

Choose a tag to compare

@github-actions github-actions released this 13 Nov 03:08
0ac0b60
Inbrowse 0.4.1

v0.4.0

Choose a tag to compare

@github-actions github-actions released this 12 Nov 19:34
53270d6
Inbrowse 0.4.0

Inbrowse 0.3.0

Choose a tag to compare

@github-actions github-actions released this 03 Nov 07:09

Inbrowse 0.3.0 adds media, OCR and research tools to the CLI and the MCP server.

New

  • inbrowse media / MCP media_extract: still frames (PNG, at an interval and at scene changes), 16 kHz mono WAV audio and captions from MP4, M4A, WebM, Matroska and WebVTT. H.264, VP8, VP9, AAC and Opus decode in software; nothing extra to install.
  • inbrowse ocr / MCP image_ocr, and --ocr / ocr: true on media frames: text from images through the OCR model in the Incord memory engine. Images too large for the engine are sent as overlapping tiles and merged, never refused.
  • MCP social_search / social_thread and inbrowse social: posts with engagement counts from Reddit, X, Hacker News, GitHub and Polymarket, ranked and de-duplicated.
  • MCP document_extract: text of PDF, Office, OpenDocument, spreadsheet, CSV and text files, as inbrowse extract prints it.

Works with Watchman

Watchman runs the Incord memory engine that serves OCR (Inbrowse calls it at http://127.0.0.1:9180 by default) and keeps the cookies and API keys the research tools sign in with. Inbrowse never stores either. Everything also works from the CLI or MCP without Watchman: set INBROWSE_OCR_URL, pass --cookies, and put API keys in environment variables.

Notes

  • Windows builds load Media Foundation and Direct3D 11 only when used, so the binary starts on Windows N editions.
  • Each archive's checksum is in SHA256SUMS.

Full list in CHANGELOG.md.