Plugin

Cervello DeepSeek + trascrizione automatica delle immagini: allega immagini nella GUI e ognuna viene trascritta tramite il deepseek-v4-flash-vision-exp ufficiale per impostazione predefinita (un cervello V4-Pro puramente testuale può vedere le immagini), con qualsiasi VLM compatibile con OpenAI o Ollama locale come alternative.

Per il modello di visione nativo di DeepSeek Harness deepseek-v4-flash-vision-exp, innalza i limiti di ammissione delle immagini a 32 MiB / 8192 px / 600 immagini e aggiunge uno strumento highres_read che suddivide le immagini grandi in riquadri, per poi restituire l'immagine intera più riquadri di 800x800 tramite lo strumento read_image dell'host.

DeepSeek Harness plugin: auto-route image-bearing requests to deepseek-v4-flash-vision-exp, then fall back to the original model.

Composer stats strip for the DSH Web UI: official CNY pricing with auto-synced peak/off-peak tiers, per-model accounting, live timers, subagent-tree merging, DeepSeek account balance, budget alerts, and streaming cost estimation.

Vision bridge for text-only DeepSeek models: transcribes attached images with deepseek-v4-flash-vision-exp before they reach deepseek-v4-pro, with no third-party dependencies.

DeepSeek balance and per-session cost in the DSH web composer dock, with official peak/off-peak pricing (peak hours Mon-Fri BJT 9-12 / 14-18, weekends off-peak) and built-in deepseek-v4-flash-vision-exp vision model rates billed at flash price (images billed as tokens, max 384 tokens each).

Vision browser tool: drives real Chrome over CDP with browser-use and reads the page with deepseek-v4-flash-vision-exp, so canvas text, text baked into images and values in rendered charts are readable; returns JSON validated against a caller-supplied schema and reports per-run token cost.

LLM-as-a-Verifier for dsh with the Best-of-N conversation mode built in: rank N candidates with a fine-grained verifier (expected grade over the logprob distribution), on demand via the verify tool or automatically on every turn of a Best-of-N session. In