ClawsGO 是下一代 AI for Science 产品，探索端到端的科研新范式

快速开始 ClawsGO 的真实研究

Genomics LabPro

New task

Hosts

Agent

Team & Subscription

Skill Market

Examples & tutorials

Pinned tasks

Projects

New projectOne directory, all related tasks

Delivery systems review3d agoCloud host

Lab notebook 20261w agoMacBook Pro

Tasks

Figure 3 — regression plots4h agoCloud host

Rebuttal draft, reviewer 2YesterdayCloud host

Protein docking batch run2d agoMacBook Pro

Grant proposal outline3d agoCloud host

Invite friends, earn creditsEarn 1,500 credits per activated friend

WWenjie Zhao

CRISPR delivery meta-analysisFigure 3 — regression plots

Pull the 2020–2026 trials comparing LNP and AAV delivery, run a random-effects meta-analysis, and draft the results section.

ClawsGO 1.5 Titanium

Searching the literature first, then I'll extract effect sizes and fit the model.

Worked through 12 steps

Searching "LNP vs AAV delivery efficiency"4.2s

Reading pubmed.ncbi.nlm.nih.gov2.8s

Fit random-effects meta-analysis18s

Writing results_draft.tex+64

LNP vs AAV delivery — results draft

results_draft.pdf

Open

forest_plot.pngPNG

Describe your task, / for commands and skills, @ for files and tasks

ClawsGO 1.5 Titanium·High

ClawsGO Science can make mistakes. Verify important information.

Overview

Steps

Collect 2020–2026 trials

Extract effect sizes

Fit random-effects model

Draft the results section

Assemble forest plot figure

受来自顶级研究机构的用户信赖

为 AI Native 科研范式而构建

无人值守

留下一个问题，合上电脑。几个小时后，交给你一份可以直接批注的完整研究。

可溯源的过程

环境、命令、文件改动、文献检索，每个动作都在它发生的位置上；整轮工作随时重放。

Worked through 9 steps

Searching "perovskite degradation"3.6s

Reading nature.com/articles/s415602.1s

Writing fit_model.py+72

python fit_model.py --studies 5224s

$ python fit_model.py --studies 52

loaded 52 studies · 1,184 device lifetimes

fitted degradation model in 23.6s

encapsulation dominates rate: β = 0.71 (95% CI 0.63–0.79)

leave-one-out check: stable across all 52

wrote results/fit.json

4:07 / 26:121x

常驻云端，三端同步

智能体在独立云主机上工作，数小时不休眠不中断。网页、手机、桌面看到的是同一个任务、同一时刻的状态。

建立在最强的模型之上

Claude Fable 5.1、Claude Opus 5.5、GPT 6 Astra 都在菜单里，按任务随手切换。云主机运行时长不另行计费——你只为智能付费。

GPT 6 Astra

OpenAI 最深的推演

Claude Opus 5.5

长程分析的主力

Claude Fable 5.1

Mythos 级旗舰

从一个问题，到一份带引用的交付物

睡前一句话，醒来一份初稿。

23:47

描述任务。

一句话就够。

调研 CRISPR 递送近五年的文献，起草一篇综述。

23:48 – 07:12

它开始工作。看不看，由你。

它检索学术信源、阅读文献、编写并运行代码、起草文稿——执行计划与进度实时可见，任何设备。

Breaking the question into a protocol: scope, sources, extraction plan.

Define inclusion criteria — 2020–2025, in vivo + clinical1m

protocol.md+38

Casting a wide net first — five sources, then dedupe.

Search PubMed — 428 records6m

Search Embase — 312 records4m

Search Web of Science — 465 records5m

Search bioRxiv preprints — 94 records3m

Search ClinicalTrials.gov — 61 trials2m

Dedupe 1,360 → 887 unique records2m

Thought for 26s

Screening in two passes: titles and abstracts first, then full text.

Pass 1: titles + abstracts — 887 → 30241m

Pass 2: full-text eligibility — 302 → 2141h 12m

Read 214 full-text papers3h 02m

screening_log.csv+887

Extracting outcomes into a structured evidence table.

evidence_table.csv+214

Normalize dosing units across 41 studies14m

python validate_extraction.py8s

Effect sizes look heterogeneous — adding subgroup and sensitivity analyses.

python meta_analysis.py12s

python subgroup_by_vector.py11s

python sensitivity_check.py9s

python publication_bias.py7s

Thought for 41s

fig1_prisma_flow.png3s

fig2_forest_plot.png4s

fig3_efficacy_by_vector.png3s

fig4_bias_funnel.png3s

Drafting section by section, citing as I go.

Outline — 7 sections+24

Draft §1 Introduction+42

Draft §2 Delivery landscape+118

Draft §3 Viral vectors+87

Draft §4 Lipid nanoparticles+102

Draft §5 Emerging systems+64

Draft §6 Clinical translation+96

Draft §7 Outlook+38

Second pass: tightening every claim against the evidence table.

Cross-check 96 citations7m

Revise §2, §4 — reconcile conflicting results+31

Verify every figure is referenced2m

Thought for 12s

latexmk review_draft.tex21s

review_draft.pdf — 38 pages, 96 citations07:12

这一夜，你在睡觉。

07:12

收取交付物。

一份真实的文件：报告、数据、图表、LaTeX 手稿。归你保存、引用、继续使用。

7h 25m · 1,360 records · 38 pages

CRISPR delivery — a five-year review

review_draft.pdf

Open

fig2_efficacy.pngPNG

它替你浏览开放网络

云端浏览器与智能体同任务、同页面。它打开论文、抽取方法，并留下随时可接管的轨迹。

Cloud browserClick to take over

PubMed

pubmed.ncbi.nlm.nih.gov/?term=solid-state+electrolyte+interface&sort=date

PubMed®solid-state electrolyte interface

1,284 results · sorted by most recent

Interfacial degradation pathways in sulfide-based all-solid-state batteries
Kato Y, et al. · Nat Energy · 2026
Interfacial resistance growth dominates capacity fade; cross-sectional imaging links void formation to local current density…
Free full text
In situ imaging of the LGPS | Li interface under fast cycling
Chen R, et al. · Joule · 2026
Interfacial resistance growth dominates capacity fade; cross-sectional imaging links void formation to local current density…
Free full text
Coating strategies for stabilizing electrolyte–cathode interfaces
Alvarez M, et al. · Adv Mater · 2025
Interfacial resistance growth dominates capacity fade; cross-sectional imaging links void formation to local current density…
Free full text

为 AI 原生打造的 LaTeX 工作台

源码与 PDF 并排，每次停顿自动编译。与智能体共写、自己精修——含 SyncTeX 与期刊引擎。

LaTeX · review.texSyncTeX · auto-compile

12\section{Results}

13Sulfide electrolytes reach $10^{-2}$\,S/cm at room

14temperature~\cite{kato2026}.

16\subsection{Ionic conductivity}

17LGPS retains higher conductivity than oxide

18garnets across the measured range, consistent

19with recent reports~\cite{chen2026}.

21\begin{figure}[t]

22 \includegraphics{fig1.figure}

23 \caption{Ionic conductivity trends}

24\end{figure}

26\subsection{Interface stability}

27Interfacial resistance decreases once the

28coating layer exceeds a critical thickness.

Compiled · 2s agoXeLaTeX

Results

Sulfide electrolytes reach 10⁻² S/cm at room temperature [12]. Among candidate solid electrolytes, LGPS remains the conductivity leader under ambient conditions.

Ionic conductivity

LGPS retains higher conductivity than oxide garnets across the measured range, consistent with recent reports [8].

Figure 1: Ionic conductivity trends across candidate electrolytes.

Interface stability

Interfacial resistance decreases once the coating layer exceeds a critical thickness, in line with the imaging study of Chen et al.

可反复修订的出版级图表

每张图都是可复现包：图、代码、环境与版本历史。圈出问题，让智能体改，保留每一次修订。

Legend clear of the curve, and larger LGPS markers.

ClawsGO 1.5 Titanium9s

Marking the overlap, then fixing each item.

Opened fig1.figure

Wrote 2 revision items

Repositioning the legend

Send a message, / for commands and skills

ClawsGO 1.5 Titanium · High

接入你的实验室 GPU

通过 SSH 注册实验室主机。智能体在你的硬件上运行重计算任务，把结果写回云端工作区。

gpu-lab · A100 80 GB ×2Running

ssh gpu-lab · ~/workspace/battery

$ ssh gpu-lab

connected · ubuntu 24.04 · cuda 12.8

$ nvidia-smi --query-gpu=name,memory.used

NVIDIA A100-SXM4-80GB, 38.1 GiB

$ uv run python train.py --epochs 40

loaded 1.2 M structures · 40 epochs planned

[epoch 26/40] loss=0.149 · gpu 82%

[epoch 27/40] loss=0.142 · gpu 83%

[epoch 28/40] loss=0.137 · eta 11 m

checkpoint saved · model-028.safetensors

syncing → workspace/runs/exp-07/

GPU 83%·38.1 / 80 GB·epoch 28 / 40Syncing

值得署上你名字的交付物

研究报告

文献综述与政策扫描，深入、带引用。

Research report12 Sep 2026 · p. 1 / 24

Solid-state electrolyte interfaces

Prepared by ClawsGO · 48 sources · 31 peer-reviewed

Summary

Sulfide electrolytes lead on room-temperature conductivity[1, 3]; oxide garnets retain the widest electrochemical window[7].

Table 1 · Room-temperature ionic conductivity

LGPSsulfide1.2 × 10⁻²

Li₆PS₅Clargyrodite3.0 × 10⁻³

LLZO (Al)garnet4.0 × 10⁻⁴

LiPONthin film2.0 × 10⁻⁶

2. Interface degradation mechanisms

Across the 31 cycling studies, capacity fade correlates with interphase growth rate rather than bulk conductivity[7, 12].

[7]Kato et al. · Nat. Energy 2016

High-power all-solid-state batteries using sulfide superionic conductors

Verified · 10.1038/nenergy.2016.30

学术论文

带真实引用的 LaTeX 草稿，编译为 PDF。

Journal of Power Sources 612 (2026) 235–242Research article

Interface stability of sulfide solid electrolytes against lithium metal anodes

L. Chen^a, M. Okafor^b, R. Sato^a,*

^a Department of Materials Science, Tohoku University · ^b Institute for Energy Storage, ETH Zürich

Abstract Sulfide electrolytes reach 10⁻² S/cm at room temperature [12], yet interfacial decomposition against lithium metal limits cycle life. We combine impedance spectroscopy with operando XPS to resolve the reaction front [8], and show that a 5 nm LiNbO₃ buffer suppresses interphase growth by 74 %.

Keywords: solid electrolyte · sulfide · interphase · operando XPS

1. Introduction

Solid-state batteries promise higher energy density and intrinsic safety, provided the electrolyte remains stable against both electrodes [1,2]. Among candidates, Li₁₀GeP₂S₁₂ (LGPS) combines the highest bulk conductivity with the narrowest stability window.

In the bulk, ionic transport follows an Arrhenius relation between conductivity and temperature,

σ(T) = σ₀ exp(−E_a / k_BT)(1)

where E_a is the activation energy. Fig. 1 compares LGPS with Al-doped LLZO across 25–100 °C.

Compiled · XeLaTeX · 0 errors · 1.2s

科学图表

由代码生成的出版级图，每版可复现。

为各种研究工作而生

科研与学术

从问题到带引用的综述报告

Microplastics and systemic inflammatory markers — working draft1

Microplastic exposure and systemic inflammatory markers in human populations

A systematic literature review of human studies (2018–2026)

Environmental health · Epidemiology · PRISMA 2020 · Literature searched 10 Aug 2026

Review profile

Protocol: MP-INFLAM-2026-01 · internal
Search date: 10 Aug 2026
Databases: EuropePMC · PubMed · CrossRef
Eligibility: Human studies 2018–2026, quantified exposure
Risk of bias: ROBINS-I adapted · 5 domains
Included: 24 reports · 11 with quantitative data

Abstract (structured)

Background. Micro- and nanoplastics are pervasive pollutants now detected in blood, faeces, urine, lung tissue and placenta [1–4]; whether internal burden is associated with systemic inflammation in humans is not established.

Objective. To identify and appraise human observational studies (2018–2026) relating a microplastic exposure proxy to inflammatory markers (CRP, IL-6, TNF-α, fibrinogen).

Results. In the largest study (n = 420), urinary MP concentration was associated with hs-CRP (β = 0.28, P < 0.001) and IL-6 (β = 0.13, P = 0.002) [7]. Carotid-plaque MNPs predicted a composite of MI, stroke or death (HR 4.53, 95 % CI 2.00–10.27) [9].

Conclusions. Internal MP burden is consistently but weakly associated with inflammatory markers — most reproducibly IL-6 — across urinary, blood and faecal matrices. The evidence base does not support causal claims.

Figure 1 · PRISMA flow

Records identifiedn = 8,803

Duplicates · 2,391

Records screenedn = 6,412

Excluded · 6,225

Full texts assessedn = 187

Excluded · 163

Studies includedn = 24

Table 1. Characteristics of included studies with quantitative exposure–inflammation data (ROB = risk of bias, ROBINS-I adapted).

StudyDesignnExposure (matrix, method)Adjusted estimateROB

Cui 2026Cross-sectional420Urine · micro-FTIR + Ramanβ hs-CRP 0.28 (P < 0.001)M

Yang 2024Cross-sectional101Blood · Pyr-GC/MSβ IL-6 0.042 (P < 0.05)M

Marfella 2024Prospective cohort257Carotid plaque · SEMHR 4.53 (2.00–10.27)M

Horvatits 2022Case–control52Faeces · µ-FTIR41.8 vs 28.0 items/gS

26 references verified · EuropePMC · CrossRef

咨询与尽调

尽职调查与客户简报，信源可核

Due diligence memoConfidential

Acme Robotics · Series C

Prepared for Northwind Capital · 14 Sep 2026 · 62 sources

Key risks

Revenue mixTop 2 customers · 61 % of ARRHigh[S3]

IP ownership3 of 9 patents held by founderHigh[S11]

Key personCTO vesting cliff · Q1 2027Med[S7]

LitigationNone open · 5-yr PACERLow[S18]

Recommendation

Proceed, conditional on curing the IP assignment before close; price at 0.85× of lead terms[S3, S11].

政策与智库

政策扫描与证据简报，逐条带出处

Policy brief · No. 14September 2026

Implementing the EU AI Act: obligations for general-purpose models

What applies when, and what it means for providers placing models on the EU market.

Application timeline · OJ L 2024/1689

2 Feb 2025Prohibited AI practices applyArt. 5
2 Aug 2025General-purpose AI model obligationsArt. 53
2 Aug 2026High-risk system requirementsArt. 6 · Annex III
2 Aug 2027Legacy GPAI models must complyArt. 111(3)

2. What providers must do

Providers of general-purpose models must maintain technical documentation, publish a training-content summary and comply with EU copyright law [Art. 53(1)].

4 / 4 claims sourced · Official Journal

金融与投研

行业地图与公司深研

工程与技术评估

技术评估、RFC 草稿、迁移计划

RFC-042Draft · v3Updated 2 h ago

Migrate authentication from SAML to OIDC

Author: platform team · Reviewers: 3 · Target: Q1 2027

Options considered

OptionEffortRiskVerdict

A · Patch existing SAML2 wkMedRejected

B · OIDC via Keycloak6 wkLow✓ Recommended

C · Vendor IdP3 wkLowDeferred · cost

Migration plan

Dual-run both providers behind a feature flag (2 wk).
Migrate service accounts and rotate client secrets.
Cut over; keep SAML read-only for 30 days, then remove.

市场与竞争研究

市场扫描与竞品跟踪

Competitor matrix · Q3 20265 products · 12 capabilities

AI writing tools — capability matrix

Verified against public pricing and docs · 10 Aug 2026

CapabilityDraftlyPenwiseScriboNotelyQuillo

Web editorYesYesYesYesYes

Docs exportYesPartialPartialYesYes

Citation supportPartialYesNoPartialNo

Team workspaceYesYesYesYesPartial

API accessYesYesPartialPartialNo

SSO / SAMLYesPartialYesYesNo

On-prem / VPCNoNoNoNoYes

BitMiracle AI 是一支致力于构建下一代 AI 科研基础设施的独立研究团队

加入我们

ClawsGO 是下一代 AI for Science 产品，探索端到端的科研新范式ClawsGO 是下一代 AI for Science 产品探索端到端的科研新范式