DeepSeek Harness Plugins · Model Inference
Plugins that extend DeepSeek Harness (DSH) with vision, workflows, memory, web tools and more.
Add the OpenAI Codex model route to DeepSeek Harness via ChatGPT subscription OAuth, no API key needed, with live model discovery.
Adds Ant Digital MaaS and NVIDIA NIM streaming model routes to a DeepSeek Harness profile with DSH-managed credentials and no extra proxy.
An enhanced model selector for DSH with a single-level search and grouping menu, inline reasoning-effort slider, full keyboard navigation, and automatic strongest-thinking tier selection for reasoning models.
AST code skeletonizer and O(n) heuristic log compressor for DeepSeek Harness agents, cutting context token consumption by up to 85% while preserving essential architectural interfaces and failure traces.
An image background removal plugin for DSH, supporting CPU or NVIDIA GPU modes with configurable models via the Web GUI.
A deterministic UTF-8 byte cap on tool result strings that returns a labelled prefix with the original sha256 digest—no model calls, no randomness.
Adds native ChatGPT OAuth sign-in to DeepSeek Harness's built-in OpenAI Codex provider by connecting the Web Models page to the host authorization service, without replacing the LLM adapter.
Translates text-based PDFs paragraph by paragraph via DeepSeek or any OpenAI-compatible API, producing an editable output PDF that preserves layout, styles, images, and links, with term injection and resumable caching.
A DeepSeek Harness model-inference plugin providing zcode-related masking capabilities.
Context compaction plugin for DeepSeek Harness featuring API and local-rule summarization engines, real-time pruning, anti-loop guardrails, and full /restore rollback without breaking append-only session logs.
A zero-token, fully offline math input plugin for DeepSeek Harness supporting handwriting recognition, screenshot OCR, and LaTeX editing, with all inference running locally in the browser.
Probes an OpenAI-compatible endpoint with five fixture requests and returns a protocol compatibility card covering streaming, token limits, and endpoint availability.