CloscoScience
An automated agent platform for scientific research scenarios. It provides persistent cloud hosts preloaded with scientific computing environments and supports multi-model collaboration and automated task execution.
Timeline (last 90 days)
CloscoScience is a scientific research automation Agent that solves the three hurdles of networking, account payment, and simulation environment by providing a team-shared cloud host.
The cloud host is configured with 8 cores and 16G of memory, nearly 500G of disk, runs Ubuntu 24.04, and comes preinstalled with more than 2,400 Python packages, including R and TeX.
In the 8-hour CFD validation task, the Agent used only 24% of its context window.
The prompt set acceptance criteria of lift deviation within 2% from CFL3D and a grid independence verification across three meshes.
The medium-mesh computation diverged in the early stage, and the Agent autonomously located pressure oscillations above the airfoil surface and completed troubleshooting by changing the numerical scheme, using an incompressible sphere solution as a comparison, and fixing the cold-start issue.
The entire task took nearly 8 hours and 197 steps, and ultimately all 6 acceptance criteria were met.
The deliverables include 5 files: a verification report, a reproducible compute package, comparison plots, and a data table.
The group's existing machines (such as a Fluent workstation or GPU cluster) can access CloscoScience by being registered as SSH servers, and a personal computer can also join the team as a host.
CloscoScience is a scientific research automation Agent that is directly usable on domestic networks, provides one cloud host per team, offers an Ubuntu 24.04 environment with more than 2,400 Python packages preinstalled, complete R and Julia toolchains, and a terminal that can be opened directly.
CloscoScience's troubleshooting capability is built on the existing solver ecosystem; for obscure self-developed software, environment configuration may still get stuck; error attribution works for known textbook problems, but unknown deviation patterns still require human judgment.
7 more timeline events
All Facts (17)
CloscoScience recommends the master-subagent mode, with the main model being Cloud Opus 5.5 (Anthropic's current latest flagship model), sub-agents following the main model, and reasoning intensity set to high.
60%UnverifiedCloscoScience's skill marketplace has 860 skills, and installing once shares them across the whole team.
60%UnverifiedIn the 8-hour CFD validation task, the Agent used only 24% of its context window.
50%UnverifiedThe prompt set acceptance criteria of lift deviation within 2% from CFL3D and a grid independence verification across three meshes.
50%UnverifiedCloscoScience is a scientific research automation Agent that solves the three hurdles of networking, account payment, and simulation environment by providing a team-shared cloud host.
50%UnverifiedThe entire task took nearly 8 hours and 197 steps, and ultimately all 6 acceptance criteria were met.
50%UnverifiedThe deliverables include 5 files: a verification report, a reproducible compute package, comparison plots, and a data table.
50%UnverifiedThe group's existing machines (such as a Fluent workstation or GPU cluster) can access CloscoScience by being registered as SSH servers, and a personal computer can also join the team as a host.
50%UnverifiedThe cloud host is configured with 8 cores and 16G of memory, nearly 500G of disk, runs Ubuntu 24.04, and comes preinstalled with more than 2,400 Python packages, including R and TeX.
50%UnverifiedCloscoScience's troubleshooting capability is built on the existing solver ecosystem; for obscure self-developed software, environment configuration may still get stuck; error attribution works for known textbook problems, but unknown deviation patterns still require human judgment.
50%UnverifiedCloscoScience's cloud host is configured with 8 cores and 16G of memory, nearly 500G of disk, and comes preinstalled with more than 2,400 Python packages, two sets of OpenFOAM, and Paraview 6.1.
50%UnverifiedCloscoScience's environment is persistent, and software packages you install yourself will not be lost due to updates.
50%UnverifiedCloscoScience supports registering a team's existing Fluent workstation or GPU cluster as an SSH server for access, and a personal computer can also join the team as a host.
50%UnverifiedIn the NACA 0012 airfoil verification task, the Agent ultimately met all 6 acceptance criteria, with a total runtime of nearly 8 hours and 197 steps, 24% context usage, and returned 5 files.
50%UnverifiedIn the report, the Agent attributed the low maximum lift to the SST turbulence model's inability to capture the sudden stall separation in the wind tunnel, which is a model limitation rather than a solver setup issue.
50%UnverifiedCloscoScience is a scientific research automation Agent that is directly usable on domestic networks, provides one cloud host per team, offers an Ubuntu 24.04 environment with more than 2,400 Python packages preinstalled, complete R and Julia toolchains, and a terminal that can be opened directly.
50%UnverifiedThe medium-mesh computation diverged in the early stage, and the Agent autonomously located pressure oscillations above the airfoil surface and completed troubleshooting by changing the numerical scheme, using an incompressible sphere solution as a comparison, and fixing the cold-start issue.
50%