AIPOCH

Open Science v0.6.0 is now available. 🎉 https://lnkd.in/gkHbibAG Open Science is expanding beyond the desktop into a more connected AI research platform. This release introduces remote compute, programmable automation, and better interoperability with existing scientific workflows. Whether you're running notebooks on a local machine, an SSH server, or integrating Open Science into your own pipelines, v0.6.0 gives researchers more flexibility without sacrificing reproducibility. 【What's new】 ✅ Remote SSH Compute Register remote servers as Notebook execution environments and switch seamlessly between local and remote kernels—laying the foundation for cloud GPUs and cluster-based research. ✅ Headless Task SDK & CLI Automate research workflows using the new zero-dependency Node.js SDK and CLI. Create tasks, monitor execution, retrieve artifacts, and reuse the same projects, sessions, permissions, and credentials from the desktop application. ✅ Export Sessions as Jupyter Notebooks Any Notebook Session can now be exported as a standard .ipynb file, individually or in batches, making it easier to share and continue work across scientific ecosystems. ✅ Desktop Task Notifications Receive system notifications when long-running Agent or Notebook tasks complete or fail—no need to keep watching the application. ✅ More Model Flexibility Added support for Volcengine Ark and configurable reasoning effort (Low / Medium / High) for supported models. Additional improvements • Deep links for Projects and Sessions in Web UI • Editable and copyable user messages • Built-in update download progress • WebSearch is back for Agents • Redesigned Agent Framework management • Improved provider compatibility testing • Better error messages and session recovery • Enhanced Codex permission isolation • Windows R detection and stability improvements Open Science continues evolving from an AI workbench into an open platform for reproducible scientific research—connecting local workflows, remote compute, and programmable automation in one environment. #OpenScience #ScientificAI #OpenSource #ResearchInfrastructure #SSH #Jupyter #BuildInPublic #AIPOCH

Open Science v0.1.1 https://lnkd.in/gkHbibAG Building research software isn't just about adding features. It's about removing friction. This release focuses on making Open Science something more researchers can actually install and use. Windows joins macOS and Linux. First-time setup guides users through configuration instead of expecting environment variables. Model providers can be configured inside the app, and local diagnostic logs help troubleshoot without sending telemetry elsewhere. Open scientific infrastructure should be accessible—not only to developers, but to researchers who simply want to start working. Highlights • Windows, macOS, and Linux support • Guided first-run setup • Configurable model providers • Secure API key storage • Local diagnostics with no default telemetry Scientific tools should spend less time asking users to configure software—and more time helping them do science. #OpenScience #ResearchSoftware #ScientificComputing #OpenSource #MedicalResearch #AIResearch #BuildInPublic #AIPOCH

Open Science v0.1.0 https://lnkd.in/gkHbibAG AI for science shouldn't end with a chat. This first public release is about proving a different interaction model: create a research project, upload your data, let an agent plan and execute multi-step analyses, inspect the generated code, and keep the entire session for later. Scientific work produces more than answers. It produces artifacts—code, notebooks, figures, tables, execution history, and decisions that deserve to be preserved. That's the foundation we're building with Open Science: an open AI research workbench where research remains reproducible long after the conversation ends. What's included • Multi-step agent planning and execution • Persistent Python kernels • Preview for CSV, FASTA, HTML, JSON, Markdown, images, and more • Recoverable projects, sessions, and execution history • Human approval before high-risk actions More to come. #OpenScience #AI #ScientificAI #OpenSource #ResearchInfrastructure #Reproducibility #MedicalAI #AIPOCH

People often describe model-agnostic as "supporting multiple models." https://lnkd.in/gkHbibAG That misses the more important idea. Research workflows tend to outlive the models that execute them. A literature review, an analysis pipeline, a verification step, or an evidence trail shouldn't need to be rebuilt every time a new model appears. - The workflow is the scientific asset. - The model is just one execution engine. That's one of the design principles behind Open Science. Build workflows that can evolve with models—not workflows that expire because of them. One scientific workflow. Any model.

Open Science is our vision for AI-native scientific infrastructure. https://lnkd.in/gkHbibAG We believe researchers should be able to choose their own models, run workflows on their own infrastructure, inspect every research skill, preserve complete provenance, and reproduce every result without being locked into a single deployment model. The emergence of AI research workbenches marks an important shift for scientific AI. Researchers increasingly need systems that can plan analyses, execute workflows, manage artifacts, and preserve the full context behind every result, not just generate text or code. Claude Science is one example of this direction, and we believe it signals an important change for the field. Our goal with AIPOCH Open Science is to make that vision more open: self-hosted, model-agnostic, community-driven, and accessible to research teams with different infrastructures, compliance requirements, and ways of working. Scientific infrastructure should be as open as science itself.

Open Science is the open answer to Claude Science: a self-hosted, model-agnostic AI research workbench for reproducible science. https://lnkd.in/g5b2jqFV The goal is not to make AI write a few scripts. The goal is to make the entire research workflow inspectable: planning, execution, artifacts, provenance, and verification. Most research work is still scattered across notebooks, RStudio, SSH clusters, database tabs, reference managers, and manuscript editors. The code lives in one place. The data version in another. Figures are exported manually. Environment details disappear. Three months later, reproducing one plot can be harder than rerunning the whole project. An AI-native workbench should connect these pieces instead of adding another chat window on top. Claude Science shows why this category matters. A useful research agent should be able to plan analyses, execute code, call scientific databases, generate figures and tables, save artifacts, check outputs, and preserve provenance. But if this layer depends on one vendor, one model family, one hosted cloud, and one subscription gate, it cannot fully serve open science. Scientific infrastructure should be inspectable, forkable, and owned by the people doing the work. Open Science is building that layer in the open. Any model. Self-hosted by default. Skills researchers can read, fork, and improve. Artifacts that stay tied to code, data, environment, logs, and lineage. The current page shows a real paper reproduction flow: PDF → agent plan → live report → saved figures/tables → concordance checks against the source paper.

Every research lab has knowledge that never makes it into the paper. https://lnkd.in/gkHbibAG - Why a cohort was filtered a certain way. - Why one preprocessing pipeline was trusted over another. - Which quality checks repeatedly failed. - Which statistical assumptions were debated before the final analysis. These decisions often live in notebooks, Slack threads, or the memory of the people who made them. We think AI should help preserve that layer of scientific knowledge—not just generate text or code. That's the vision behind AIPOCH Open Science: an open ecosystem where research workflows, methodological decisions, and domain expertise become reusable, inspectable, and continuously refined by the research community. Because science advances through knowledge that others can build upon.

Lab handoff is going to become a product surface. https://aipoch.com A PhD student leaves. A collaborator joins. A dataset gets updated. The analysis still has to carry the decisions that made the result believable: cohort filters, missingness rules, parameter limits, failed checks, and the small method note that says why. Most software treats that as context outside the run. We think the next serious layer for AI-assisted research is built around handoff: making enough of the method visible that another researcher can rerun, inspect, and challenge the work without rebuilding the whole story from screenshots and old notebooks. That is the kind of industry we want AIPOCH to push toward. #medical #ai #agent #science #research

Scientific progress has been built on one principle: knowledge grows when it can be shared. https://lnkd.in/gkHbibAG Papers are published. Methods are documented. Code is open-sourced. Datasets are reused. Every contribution gives the next researcher a stronger starting point. AI should follow the same principle. Today, most AI tools help individual researchers work faster. They summarize papers, write code, explain concepts, or answer questions. These capabilities are valuable, but much of the research process still remains locked inside isolated conversations. The reasoning behind an analysis, the workflow used to clean data, the parameters chosen for a model, or the sequence of decisions that led to a result are often difficult to inspect, reproduce, or build upon. We believe AI can do more than accelerate research. It can become part of the scientific infrastructure itself. That's the vision behind AIPOCH. Instead of treating AI as a chatbot, we're building an open ecosystem of reusable medical research skills. Each skill represents a validated research workflow that can be executed, evaluated, improved, and combined with others. Rather than starting from scratch for every project, researchers can build on workflows that have already been refined by the community. This is why we've developed more than 550 medical research skills covering literature review, evidence synthesis, bioinformatics, statistical analysis, study design, manuscript preparation, and many other research tasks. These skills are not isolated prompts—they are modular building blocks that can evolve alongside advances in science. Equally important, we believe research AI should be transparent. Scientific workflows should be inspectable. Evaluation should be reproducible. Methods should be shared rather than hidden behind interfaces. Open Science is not simply about open-source code; it's about making research knowledge itself reusable. The future of AI in science isn't only about building more capable models. It's about building an ecosystem where researchers can contribute workflows, improve methods, validate results, and collectively accelerate discovery. That's the future we're building at AIPOCH.

Most AI research demos stop at the script. At https://aipoch.com, we are building around what happens before and after that script: data cleaning choices, assembled analysis code, workflow steps, failed checks, and the small method notes that explain why the output changed. Because in research, a result is only useful if the path can be inspected again.