DeepSeek Harness Plugins · Model Inference
Plugins that extend DeepSeek Harness (DSH) with vision, workflows, memory, web tools and more.
Lets text-only models read TIFF/TIF images by decoding a page to PNG, reporting full header metadata, and optionally invoking a vision model to describe the page.
A Reasonix-style cache-aware compaction backend for DSH, delivering single-trigger, structurally checkpointed context compression with one summarizer call per transaction.
Per-message TTS playback, auto-announce, and voice input with dual-cloud-engine support (DashScope / Volcengine) and graceful fallback to system voices.
Freely adds Agnes omni-model vision and image generation to DSH agents, enabling in-chat image analysis and text-to-image or image-to-image workflows without hardcoded keys.
Auto-fills reasoning effort, context window, max tokens, and image modality for custom provider models using models.dev data.
Installs an agent preset that bootstraps each DSH session on the Minimal RL-trained condition, then auto-promotes to the full tool catalog after the first durable tool call.
A DSH web plugin adding a voice-practice mode: the model replies in short spoken-style sentences, the browser reads aloud via edge-tts, and you dictate with the microphone for bilingual listening and speaking drills.
Offline OCR plugin powered by Apple's macOS Vision framework, extracting text from screenshots and document images locally with no network or API key required.
Local image and document analysis plus image generation for agent harnesses, powered by OpenAI Codex CLI with zero runtime dependencies.
A visual, manual model and reasoning-effort picker for DeepSeek Harness in-process subagents, applying a deterministic per-session override visible beside the main model seat.
A floating pill widget that shows the current peak or valley time window with a countdown, supporting customizable time ranges persisted in DSH settings and auto-refreshing every 30 seconds.
Unified multi-provider image generation and editing plugin for DeepSeek Harness, supporting OpenAI Images-compatible and Google Gemini Interactions APIs.