1692775560/dsh-Mimir-Academic-research

Mimir — 一站式科研工作台插件：LaTeX 论文边写边编译、arXiv 文献管理、实验追踪、指标图表、GPU 服务器 SSH 任务编排，管理科研全周期。An open-source research workbench plugin for the whole research cycle.

What it solves

Mimir provides a unified, AI-driven workbench for the entire academic research lifecycle. It eliminates the friction of switching between disparate tools for literature review, experiment tracking, LaTeX writing, GPU cluster management, and presentation preparation.

How it works

It operates as a plugin for the DeepSeek Harness (dsh), integrating a web-based workbench with an AI agent. The agent is equipped with specific research tools (slash commands) and bundled skills to automate academic workflows. The system synchronizes data in real-time via Server-Sent Events (SSE), allowing the agent's edits to appear instantly in the user's view.

Who it’s for

Academic researchers and students who need a centralized environment to manage their papers, experiments, and remote compute resources while leveraging AI to accelerate writing and literature analysis.

Highlights

  • Nine-View Workbench: Dedicated interfaces for paper writing (Overleaf-style LaTeX), literature libraries (arXiv/web search), experiment metrics, figure organization, meeting decks, GPU server monitoring, and a research ledger.
  • Agentic Research Tools: Built-in slash commands for generating research ideas, planning, reviewing, and compiling LaTeX papers.
  • Integrated Compute Management: Tools to list, check, and submit jobs to remote GPU fleets directly from the workbench.
  • Academic Automation: Bundled skills for novelty checks, citation audits, and "de-AI" polishing of bilingual text.
  • Venue Tracking: A built-in catalog of CCF conferences and journals with deadline countdowns.

Related

  • Project
    vueuse/skills

    A collection of agent skills that enables AI agents to use the VueUse library more accurately and efficiently by reducing token consumption and hallucinations.

  • Project
    stanford-star/relbench

    RelBench is an open benchmark for predictive machine learning on relational databases, providing a standardized way to load and evaluate models on multi-table data.

  • Project
    TensorOpsAI/LLMstudio

    A prompt engineering platform that provides a unified proxy for multiple LLM providers and local models, featuring a playground UI and monitoring tools.

  • Project
    aidatatools/ollama-benchmark

    A cross-platform CLI tool that measures the tokens-per-second throughput of local LLMs running via Ollama based on the user's available system RAM.