AI Agent Hub
Back to plugins
🤖

dsh-tool-read-tiff

Model Inference Updated 2026.08.25

Run the following command in DeepSeek Harness:

dsh plugin install fulander0301/dsh-tool-read-tiff

Paste the following prompt into your AI chat to install this plugin:

Install the plugin in DeepSeek Harness by running dsh plugin install fulander0301/dsh-tool-read-tiff; the source is available at https://github.com/fulander0301/dsh-tool-read-tiff

About this plugin

TIFF is the go-to wrapper for enterprise scans, faxes, and scientific rasters, yet it is essentially invisible to text-only LLMs: browsers cannot open it and general-purpose vision pipelines do not yield text out of the box. dsh-tool-read-tiff closes that gap, letting models such as DeepSeek V4 read TIFF files directly inside a conversation without the user manually screenshotting or converting.

Each call takes a local path or http(s) URL plus a zero-based page index. The plugin decodes the target page to a viewable PNG via sharp (libvips + libtiff) and returns full TIFF header metadata: dimensions, page count, compression scheme (LZW, Deflate, PackBits, CCITT Group 3/4, JPEG, none), photometric interpretation, per-channel bit depths, byte order, and whether the file is classic or BigTIFF. Multi-page documents can be iterated page by page. When an OpenAI-compatible vision endpoint is configured, the decoded PNG is forwarded to that endpoint and the model's textual description is returned in-conversation, while the raw TIFF and PNG bytes never enter the session log. Even without a vision endpoint the tool still decodes, converts, and reports metadata, handing back a convertedPath that downstream tools can pick up.

Ideal for engineers and knowledge workers who regularly handle scanned documents, multi-page faxes, legacy reports, or scientific imagery inside a conversation flow—especially teams built around text-only models that do not want to stand up a separate image pipeline.

Use Cases

  • Extract text and tables from scanned documents and faxes within a conversation
  • Iterate through multi-page TIFFs to retrieve per-page dimensions, compression, bit depth, and other metadata
  • Send decoded TIFF pages to a vision model for structured descriptions that downstream tools can consume

Best For

  • Engineers using text-only LLMs who want to avoid building a separate image pipeline
  • Knowledge workers who handle scanned documents, multi-page faxes, or scientific rasters daily
  • Developers whose workflow is conversation-based and needs zero-config TIFF access