dsh-vision-no-vision
A DeepSeek Harness (dsh) plugin that gives a text-only LLM vision capability!!!
Compatibility
This release targets the DSH 0.2.x plugin API, and is built and tested
against dsh 0.2.1-rc.1 (Cordis 4.0.5). The manifest declares
engines.dsh: >=0.2.1-rc.1 <0.3.0, so the launcher refuses to load it on an
incompatible runtime instead of failing at call time.
Peer packages: @deepseek-ai/cordis ^4.0.5,
@deepseek-ai/dsh-tools ^0.2.1-rc.1, @deepseek-ai/dsh-llm ^0.2.1-rc.1.
Install
pnpm dsh plugin --profile web add dsh-vision-no-vision
Configuration
- name: dsh-vision-no-vision
config:
# Cap for one complete analysis (conversion + model call), in ms.
timeoutMs: 3600000
# Optional output cap; defaults to the route's advertised maximum.
maxTokens: 4096
# Optional explicit route; defaults to the session's active model.
provider: deepseek
model: deepseek-chat
Without provider/model, the tool reads the route from the calling
session's latest request header, so the image is analyzed by the same model
that is driving the conversation.
Note
Seriously, it is not a serious plugin 😝. The tool generates an ascii-art from the input image and feed it into the text-only LLM to guess the content.
Development
pnpm install
pnpm typecheck
pnpm build
The keyless smoke test runs the real tool pipeline and LLM runtime against a mock adapter:
cd smoke
node --import tsx <deepseek-harness-checkout>/vendor/cordis/bin.js
No comments yet. Be the first to write one.