Open-Science v1.2.0 sends approved jobs to four remote backends and lets imported research be questioned step by step
Synopsis
AIPOCH's local-first research workbench Open-Science released v1.2.0, letting a session submit approved jobs to Kubernetes, Hugging Face Jobs, Modal and Ray, where remote compute previously covered only SSH and Slurm. Each external submission freezes its inputs and asks for approval of target, command, resources and outputs, and recovery records let result collection resume after a restart. A built-in Open-Science Agent becomes the default for new installs, and an imported .science package or replay step can carry a child discussion without modifying the import.
Interpretation
A session can submit approved jobs directly to Kubernetes, Hugging Face Jobs, Modal and Ray alongside the existing SSH and Slurm, keeping the job and its recorded approval in the same session. Remote compute previously covered direct SSH and Slurm only, so jobs meant for these four targets could not be submitted from inside a session. Each external submission freezes its inputs and asks for approval of the target, command, resources and outputs; recovery records let result collection resume after a restart, and published results open from Compute Preview.
New installations ship with a built-in Open-Science Agent runtime as the default, so a session can start without choosing an agent first. New setups previously started with Claude as the default agent rather than a runtime shipped inside the app. The runtime uses provider-neutral turn execution, validated tool handling, streaming adapters for major API protocols and durable stores; Claude Code, OpenCode, Codex, CodeBuddy and Pi remain explicitly selectable.
Clicking Ask about this step on an imported .science package or a replay step opens a child conversation linked to that step while the parent keeps its identity. An imported package is inspectable and referenceable but stays read-only, and step-linked child discussions were not available on an imported package or a replay step. Discussions survive archiving, deleting the parent converts them into ordinary conversations, and Fork provides a writable continuation; replay reproduces recorded steps rather than re-executing them.
Perspective
These capabilities target researchers using Open-Science on macOS, Windows and Linux; remote backends run on user-provisioned targets and need the account, software, resources and permissions of the approved target, and jobs on cloud backends may incur costs on the connected account.
The release notes state that the tool does not validate scientific claims, citations or methods; replay reproduces recorded steps rather than re-executing them, solver-exact equivalence remains open, and replayable verification remains the deterministic check. Provider choice is constrained by the active framework's endpoint compatibility, and there is no local GPU compute backend or multi-user real-time collaboration.