paperfit
PaperFit is a Claude Code plugin for LaTeX academic papers. It compiles a paper, renders its pages, diagnoses layout defects visually such as wasted space, misplaced floats and overflow, and repairs the source.
Runs an end-to-end autonomous AI research pipeline, from literature search and hypotheses through experiment code, GPU cluster training, and figures to a LaTeX paper built from the real experiment results.
Claude Code
$ npx -y skills add OpenRaiser/NanoResearch --agent claude-code
Codex
$ npx skills add OpenRaiser/NanoResearch -a codex
Cursor
$ npx skills add OpenRaiser/NanoResearch -a cursor
Gemini CLI
$ npx skills add OpenRaiser/NanoResearch -a gemini-cli
GitHub Copilot
$ npx skills add OpenRaiser/NanoResearch -a github-copilot
OpenCode
$ npx skills add OpenRaiser/NanoResearch -a opencode
Windsurf
$ npx skills add OpenRaiser/NanoResearch -a windsurf



端到端自主 AI 科研引擎 — 从研究想法到完整论文,全程自动化
快速开始 · 效果展示 · 流水线 · Claude Code · 飞书机器人
🔬 NanoResearch 真正运行计算实验——它不仅生成代码,还能将代码提交到 GPU 集群执行训练,收集真实实验结果,生成论文配图,最终输出一篇有实验数据支撑的完整 LaTeX 论文。论文中的每一个数据、表格、图表都来自实际运行的实验结果,而非 LLM 编造。
以下为 NanoResearch 自动生成论文中的真实配图,所有数据、曲线、表格均来自实际运行的实验结果,非 LLM 虚构。
实验结果展示
方法对比 / 主结果

消融与可视化
以上配图均为流水线自动生成,数据来源于真实训练日志与实验结果。
NanoResearch 命令行(CLI) 提供 TUI 全屏界面 与 传统流式日志 两种呈现方式。下方为 CLI 端 TUI 主题与界面 演示视频(配色切换与布局优化)。其他入口(如 Claude Code、飞书机器人)见文档对应章节。
| 特性 | 传统 AI 写作工具 | NanoResearch |
|---|---|---|
| 文献检索 | 部分支持 | ✅ OpenAlex + Semantic Scholar 自动检索 |
| 实验设计 | ❌ | ✅ 自动生成实验方案 |
| 代码生成 | 部分支持 | ✅ 完整可运行的实验代码 |
| GPU 实验执行 | ❌ | ✅ 本地 / SLURM 自动训练 |
| 结果分析 | ❌ | ✅ 解析真实训练日志 |
| 论文配图 | ❌ | ✅ 基于真实数据 |
| 论文撰写 | 大纲/草稿 | ✅ 完整 LaTeX 论文 |
| 断点续跑 | ❌ | ✅ 任意阶段可恢复 |
| 多模型协作 | 单一模型 | ✅ 按阶段路由 |

CLI 支持 传统流式日志(非 TUI) 与 全屏 TUI 面板 两种呈现方式,可按习惯切换。



框架概览

生成论文示例

主结果(真实实验数据)

消融实验
Research Topic
↓
IDEATION → PLANNING → SETUP → CODING → EXECUTION → ANALYSIS → FIGURE_GEN → WRITING → REVIEW
↓
Exported: paper.pdf / paper.tex / references.bib / figures / code / data
📋 各阶段详细说明
| Stage | 功能 | 说明 |
|---|---|---|
IDEATION | 文献检索与创意生成 | 搜索学术文献、发现研究空白、提出假说、收集必引文献 |
PLANNING | 实验方案设计 | 将研究想法转化为详细的实验蓝图(数据集、基线、指标、消融) |
SETUP | 环境准备 | 准备代码仓库、依赖环境、模型和数据集 |
CODING | 代码生成 | 生成完整可运行的实验项目(含训练脚本、数据处理、模型定义) |
EXECUTION | 实验执行 | 在本地 GPU 或 SLURM 集群上运行训练,支持自动重试和调试 |
ANALYSIS | 结果分析 | 解析训练日志和指标,生成结构化实验证据 |
FIGURE_GEN | 图表生成 | 创建架构图、结果对比图、消融实验图 |
WRITING | 论文撰写 | 基于实验证据和引用撰写 LaTeX 论文 |
REVIEW | 审稿与修订 | 自动审阅各章节,检测问题并修订 |
🚀 EXECUTION 阶段核心能力
EXECUTION 阶段是 NanoResearch 的核心差异化能力:
遵循以下步骤,约 5 分钟即可完成从安装到首次运行的完整流程。
git clone https://github.com/OpenRaiser/NanoResearch.git
cd NanoResearch
pip install -e ".[dev]"
Tip 创建
~/.nanoresearch/config.json,替换base_url和api_key为你自己的 OpenAI 兼容 API 端点。
查看完整配置示例
{
"research": {
"base_url": "https://your-openai-compatible-endpoint/v1/",
"api_key": "your-api-key",
"template_format": "neurips2025",
"execution_profile": "local_quick",
"writing_mode": "hybrid",
"max_retries": 2,
"auto_create_env": true,
"auto_download_resources": true,
"ideation": { "model": "your-model", "temperature": 0.5, "max_tokens": 16384, "timeout": 600.0 },
"planning": { "model": "your-model", "temperature": 0.2, "max_tokens": 16384, "timeout": 600.0 },
"code_gen": { "model": "your-model", "temperature": 0.1, "max_tokens": 16384, "timeout": 600.0 },
"writing": { "model": "your-model", "temperature": 0.4, "max_tokens": 16384, "timeout": 600.0 },
"figure_prompt": { "model": "pro/gpt-5.5", "temperature": 0.5, "max_tokens": 4096, "timeout": 300.0 },
"figure_code": { "model": "pro/gpt-5.5", "temperature": 0.1, "max_tokens": 16384, "timeout": 600.0 },
"figure_gen": {
"model": "gpt-image-2",
"image_backend": "openai",
"base_url": "https://your-image-endpoint/v1/",
"api_key": "your-image-api-key",
"temperature": null,
"timeout": 600.0
},
"review": { "model": "your-model", "temperature": 0.3, "max_tokens": 16384, "timeout": 300.0 }
}
}
环境变量覆盖:NANORESEARCH_BASE_URL / NANORESEARCH_API_KEY / NANORESEARCH_TIMEOUT
# 验证配置
nanoresearch run --topic "Adaptive Sparse Attention Mechanisms" --dry-run
# 启动完整流水线
nanoresearch run --topic "Adaptive Sparse Attention Mechanisms" --format neurips2025 --verbose
# 从断点恢复(若某阶段失败)
nanoresearch resume --workspace ~/.nanoresearch/workspace/research/{session_id} --verbose
# 导出论文
nanoresearch export --workspace ~/.nanoresearch/workspace/research/{session_id} --output ./my_paper
完成流水线后,你将得到包含真实实验数据的论文配图与 LaTeX 源码。
evo 是 NanoResearch 的自进化流水线,面向个性化科研自动化。它围绕技能演化、记忆演化和反馈驱动的 planner / router 适配运行,让系统在多轮研究中复用经验并逐步贴合用户偏好。
evo# 1. 初始化或更新用户画像、偏好和运行配置
nanoresearch init
# 2. 启动自进化完整流水线
nanoresearch run --pipeline evo --topic "your research topic" --format neurips2025 --verbose
# 3. 中途失败或断开后恢复
nanoresearch resume --workspace ~/.nanoresearch/workspace/research/{session_id} --verbose
# 4. 查看阶段状态和产物
nanoresearch status --workspace ~/.nanoresearch/workspace/research/{session_id}
nanoresearch inspect --workspace ~/.nanoresearch/workspace/research/{session_id}
# 5. 导出最终论文包
nanoresearch export --workspace ~/.nanoresearch/workspace/research/{session_id} --output ./paper_export
evo 仍然运行 9 个阶段:IDEATION -> PLANNING -> SETUP -> CODING -> EXECUTION -> ANALYSIS -> FIGURE_GEN -> WRITING -> REVIEW。区别在于它会在这些阶段之间持续更新和复用用户画像、技能库、项目记忆和反馈路由。
关键机器可检查 artifact 通常位于:
~/.nanoresearch/workspace/research/{session_id}/experiment/configs/experiment_matrix.json
~/.nanoresearch/workspace/research/{session_id}/experiment/results/metrics.json
~/.nanoresearch/workspace/research/{session_id}/experiment/results/run_manifest.json
~/.nanoresearch/workspace/research/{session_id}/experiment/results/final_metrics.json
~/.nanoresearch/workspace/research/{session_id}/experiment/results/optimization_history.csv
~/.nanoresearch/workspace/research/{session_id}/experiment/results/pareto_front.json
写作阶段只读取真实执行结果、分析报告和图表 artifact。若某类实验或指标没有真实产物,系统会降低证据范围或写入 limitation / future work,不会补 synthetic 数值。文献检索可匿名使用 OpenAlex;如需更高速率,可自行配置 OPENALEX_API_KEY。
Evo/RAM 运行产生的完整 (x, y, feedback) 三元组会保存在 ~/.nanoresearch/ram_data/。安装可选训练依赖后,可训练或续训 RAM LoRA adapter:
pip install -e ".[sdpo]"
nanoresearch ram-train \
--model Qwen/Qwen2.5-7B-Instruct \
--data-root ~/.nanoresearch/ram_data \
--output ~/.nanoresearch/adapters/ram-sdpo
训练完成后,将配置中的 ram_checkpoint_path 指向输出目录即可在后续 Evo 运行中加载该 adapter。
| Stage | 任务 | 推荐模型 | 经济型 |
|---|---|---|---|
ideation | 文献检索 + 假说生成 | DeepSeek-V3.2 | DeepSeek-V3.2 |
planning | 实验设计 | Claude Sonnet 4.6 | DeepSeek-V3.2 |
code_gen | 代码生成 | pro/gpt-5.5 | DeepSeek-V3.2 |
writing | 论文撰写 | pro/gpt-5.5 / Claude Sonnet 4.6 | DeepSeek-V3.2 |
PaperFit is a Claude Code plugin for LaTeX academic papers. It compiles a paper, renders its pages, diagnoses layout defects visually such as wasted space, misplaced floats and overflow, and repairs the source.
See 12 nanoresearch alternatives
Fair-code platform to build and deploy AI agents and workflows. Combine a visual canvas with custom code, run it self-hosted or in the cloud, and connect to 1500+ integrations. AI automation you can trust with real work, from prototype to production.
A browser automation plugin for Claude Code with skills for direct browser control, scraping, testing, and QA scoring of websites, plus guidance on the Browser Use cloud browser API and Python library.
Organizes your AI agents into a team that runs around the clock, handling hiring, scheduling, delegation and reporting so you can stay in charge without staying online.
A browser automation command line tool for AI agents, letting them navigate pages, fill forms, click buttons, take screenshots and extract data from websites.
Install it with the command on its page. It includes nanoresearch-experiment, nanoresearch-ideation, nanoresearch-planning. Its skills do not fire on their own yet. Free and open source.
Is this plugin yours?
Claim it with GitHubSubmit a pluginPromote it