Dsh Llm Deepseek Vision
NagasakiSoyo-ui/dsh-llm-deepseek-vision · model-adapters
To the pure text model of DeepSeek Harness, a visual "eye" plugin was added.
★ 2GitHub stars
0Forks
2026-08-25Last updated
TypeScriptLanguage
MITLicense
Key features
- Requests without pictureszero costs, directly passing through text reasoning path (default opencode-go).
- The visual description is regenerated every round: if the session log contains a picture, each round is re-presented.
- Replaying a visual wheel reruns the visual model.
- Visual models must be capable of interpretation by their provider router
- A p-ai route entry with a handwritten declaration, if it does not declare its ability to reason, is treated as "closable thinking" (send off).
Requirements
- Yes, pi-ai route (llm-pi-aiparagraphs of settlings.yaml), and a photo modulation for model entries
- The visual model provider must be capable of authentication (configuring its apiKeyEnv in the certificate library).
Install command
dsh plugin --profile web add github:NagasakiSoyo-ui/dsh-llm-deepseek-vision