Download Latest Version aipoch-open-science-0.10.0-win-x64-setup.exe (175.8 MB)
Email in envelope

Get an email when there's a new version of Open Science

Home / v0.9.0
Name Modified Size InfoDownloads / Week
Parent folder
aipoch-open-science-0.9.0-linux-x86_64.AppImage 2026-07-31 241.8 MB
0
aipoch-open-science-0.9.0-mac-arm64.dmg 2026-07-31 221.8 MB
0
aipoch-open-science-0.9.0-mac-arm64.zip 2026-07-31 221.6 MB
0
aipoch-open-science-0.9.0-mac-x64.dmg 2026-07-31 229.6 MB
0
aipoch-open-science-0.9.0-mac-x64.zip 2026-07-31 229.5 MB
0
aipoch-open-science-0.9.0-win-x64-setup.exe 2026-07-31 175.4 MB
1 1 weekly downloads
aipoch-open-science-0.9.0-win-x64-setup.exe.blockmap 2026-07-31 173.9 kB
0
aipoch-open-science-0.9.0-win-x64.zip 2026-07-31 243.2 MB
1 1 weekly downloads
aipoch-open-science_0.9.0_amd64.deb 2026-07-31 177.0 MB
0
arm64-mac.yml 2026-07-31 375 Bytes
0
latest-linux.yml 2026-07-31 582 Bytes
0
latest-mac.yml 2026-07-31 543 Bytes
0
latest.yml 2026-07-31 383 Bytes
0
SHA256SUMS.txt 2026-07-31 850 Bytes
0
x64-mac.yml 2026-07-31 371 Bytes
0
README.md 2026-07-31 20.2 kB
0
v0.9.0 source code.tar.gz 2026-07-31 5.5 MB
0
v0.9.0 source code.zip 2026-07-31 6.3 MB
0
Totals: 18 Items   1.8 GB 2

Open Science v0.9.0

Define personal specialist agents with scoped capabilities, manage permissions with revocable grants, export entire conversations, and see exactly how many tokens each turn consumed — while DeepSeek's flash model goes native on the Responses API and framework switching, Windows upgrades, and the shared RPC contract get considerably harder.

Open Science is an open-source, model-agnostic AI workbench for scientific discovery — a self-hosted desktop app pairing a plan-and-execute agent with persistent compute, durable project/session storage, and inspectable research artifacts.

v0.9.0 is a feature release on top of v0.8.1. It brings the first concrete slices of two roadmap phases into shipped form: personal specialist agent profiles with scoped capabilities (Phase 3's "specialist roles"), and durable scoped permission management with global, project, and session-scoped allow grants (Phase 4's "scoped permissions"). Alongside those structural additions, it adds portable conversation and artifact export, per-turn token usage visibility, TIFF image previews, collapsible side panels, and native Responses API support for DeepSeek's flash model — while hardening conversation exports, Windows in-place upgrades, runtime framework isolation, and the shared RPC contract between processes.

✨ Highlights

  • Personal specialist agents. Define reusable agent profiles with their own name, avatar, persona, and scoped capabilities — a full-access catalog or a symmetric skills/connectors whitelist — and bind one per session so the identity applies from the very first turn. Existing sessions can hot-switch specialists mid-conversation behind a reconfigure barrier that fails closed (draft preserved, no partial turn on failure). (#530)
  • Scoped permission management. Tool-call approvals are no longer tied to a single conversation or an ACP-specific lifetime. Remembered allow grants now persist across global, project, and session scopes with qualifier matching, optimistic-concurrency revision checks, cascade cleanup, and runtime reconciliation — all managed from a Settings panel with filtering, per-row and family revoke, and an independently operable Undo stack. (#533)
  • Export conversations as Markdown and PDF. A whole conversation can now be exported as a portable Markdown or PDF document through a normalized public export model (not the internal session JSON), with PDF rendered in an isolated hidden window under a strict CSP. Export is disabled for empty, running, or permission-waiting sessions and is desktop-only. (#531)
  • Selective session-artifact downloads. Download all generated files from a session at once, or pick a subset through a selection dialog that loads every page, repairs an incomplete index, and saves with collision-safe naming and per-file failure reporting. (#534)
  • Per-turn token usage. Each completed agent turn now shows a compact input / cache / output token footer under the final message, with cache reads and writes combined into a single cache total. The app-managed Codex ACP adapter was patched to accumulate every model request in the turn so tool-bearing turns are no longer underreported. (#540, [#553])
  • Native Responses API for DeepSeek flash. DeepSeek's deepseek-v4-flash now routes through the native Responses endpoint instead of the Chat Completions bridge, while deepseek-v4-pro stays bridged until DeepSeek ships Responses support for it. The resolution is model-aware at a single choke point, so all downstream decisions — validation, Codex config, UI compatibility — follow automatically. (#544)
  • TIFF image rendering. The image preview now handles TIFF files alongside the existing formats, with the same zoom and pan controls. (#529)
  • Collapsible side panels. The workspace's left and right side panels can be collapsed to reclaim horizontal space, so a wide transcript or preview gets the room it needs without losing the panels entirely. (#523)

🚀 New Features

  • Personal specialist agent profiles — user-defined agent profiles with scoped capabilities (full access or a skills/connectors whitelist), session binding from the first turn, mid-conversation hot-switching behind a reconfigure barrier, and unified identity application across Claude Code (preset append + skill whitelist), Codex, and OpenCode (per-turn prompt prefix). (#530)
  • Scoped permission management — stable broker identities for persistable mutations, MCP tools, execution, file operations, and skill operations; one PermissionGrant table with global, project, and session scopes; central resolution of remembered grants across ACP, Connector, and Compute surfaces; broad-scope confirmation for project and global approvals; and a Settings panel with filtering, family revoke, monotonic optimistic updates, and a visible Undo stack. (#533)
  • Conversation export as Markdown and PDF — whole-conversation export through a normalized public model, rendered as Markdown or PDF (in an isolated hidden window under CSP), with byte-length-bounded filenames and complete-prompt title preservation. (#531)
  • Selective session-artifact downloads — a per-session Download all artifacts action with a selection dialog, batch saving to one destination folder, collision-safe naming, and per-file failure reporting without dropping successful downloads. (#534)
  • Per-turn token usage — provider-neutral input, cache, and output totals attached to the final agent message of each prompt, persisted in the session, and rendered as a compact footer; the Codex ACP adapter accumulates every model request in the turn. (#540, [#553])
  • Native Responses API for deepseek-v4-flash — a per-model Responses allow-list drives model-aware endpoint resolution so a single vendor entry can route some models natively and others through the bridge. (#544)
  • TIFF image rendering — the image preview renderer now accepts TIFF files. (#529)
  • Collapsible side panels — the workspace left and right panels collapse and expand without remounting. (#523)

🔧 Improvements

  • Public product documentation received an SEO/GEO polish pass and had the stale Vision section removed, so the landing page and README stay aligned with the shipped product surface. (#541, [#543])
  • Custom-provider Claude Code sessions no longer fire a WebFetch preflight request, removing a spurious network round-trip that could fail before the real turn starts. (#538)
  • Project deletion now surfaces failures instead of silently swallowing them, so a cache or database error during cleanup is visible and retryable. (#542)
  • Windows path-portability checks in CI are now deterministic and gated on pull requests, balancing PR speed against release validation. (#537, [#550])

🐛 Bug Fixes

  • Conversation export safety — export is hardened against incomplete session state, normalized filenames respect Unicode character boundaries, and the PDF window enforces a strict CSP without loading remote resources. (#539)
  • Windows in-place upgrade recovery — the update flow recovers when a Windows upgrade leaves the app in an unstartable state, so a failed auto-update no longer requires a manual reinstall. (#527)
  • Runtime framework switching isolation — switching the active agent framework backend at runtime no longer leaks tool state, permissions, or MCP server registrations from the previous framework into the new session. (#517)
  • Stale artifact session saves — a session can no longer persist a stale artifact reference after the artifact's owning session is deleted or migrated, preventing dangling pointers in the conversation graph. (#519)
  • Specialist validation and main-agent switching — specialist profile validation is tightened (disabled, deleted, or corrupt profiles fail closed), and switching from a specialist back to the main agent no longer carries scoped capabilities forward. (#535)
  • Shared RPC contract enforcement — the renderer-to-main RPC channel now enforces a shared type contract at the boundary, so an mismatched or stale call fails with a clear error instead of silently producing undefined behavior. (#547)
  • Undo stack positioning — the permission management Undo stack no longer overflows the window bottom edge, keeping every action reachable. (#552)
  • Advisory test stability on Windows — the advisory full test suite is stabilized against flaky Windows-specific timing and path assumptions. (#548)

📦 Install

Requirements: macOS 12+ (Apple Silicon or Intel), Linux x64, or Windows 10/11 x64. On first run, the onboarding wizard checks the environment and can install and configure an app-managed agent runtime. Once installed, the app can update itself in place.

Download the appropriate package from the Assets section below:

Platform Package
macOS (Apple Silicon) DMG for ARM64
macOS (Intel) DMG for x64
Linux AppImage or Debian package for x64
Windows Installer for x64

macOS — first launch. Official release builds are Developer ID signed and notarized by Apple, so they open like other trusted applications. A locally built copy is not notarized and may require approval through macOS Privacy & Security.

Windows — first launch (unsigned build). No Authenticode certificate yet, so SmartScreen shows a bypassable "unrecognized app" prompt (More info → Run anyway). Verify that the package came from the official release page before continuing.

Build from source instead:

npm install
npm run build:mac   # or: build:linux / build:win

Downgrading. v0.9.0 is a data-compatible release; installing an older application over a newer data directory does not require a rollback command. To prepare an isolated, compatible copy for Open Science 0.7.3, quit the app and run:

open-science rollback-to-0.7.3 --yes

The command preserves newer data and activates a separate rollback data location; it does not rewrite the current state.

🧭 What's in this release (maturity)

  • ✅ Implemented: a local-first desktop, localhost-web, headless, CLI, and task-SDK surface over persistent projects and sessions with selectable message branches; selectable Claude Code, OpenCode, and Codex agent frameworks; multi-provider model configuration with combined model/reasoning selection and native Responses API support for DeepSeek flash; persistent Python/R/REPL kernels and remote SSH execution; immutable artifact versions with inspectable audit evidence; multi-format previews including TIFF; file-based skills; permissioned scientific connectors; opt-in review and bounded correction; personal specialist agent profiles with scoped capabilities; scoped permission management with global, project, and session grants; conversation export as Markdown and PDF; selective session-artifact downloads; per-turn token usage; collapsible side panels; configurable storage; and installers for macOS, Windows, and Linux.
  • 🚧 Partial: R remains managed-only; provider choice remains constrained by the active framework's endpoint compatibility; remote compute is SSH-only; skills remain local; specialist roles are personal-only (no cross-machine sharing); and review is opt-in and record-scoped.
  • 🗺️ Roadmap: a unified model gateway, a public skills commons and cross-machine specialist sharing, Slurm/cloud-GPU execution, stronger sandboxing and credential isolation, and collaborative research workflows.

🐢 Known Limitations

  • R is managed-only. A bring-your-own R interpreter path is not built yet.
  • Remote compute is SSH-only. Slurm, cloud GPU, and parallel sub-agent fan-out are not built yet.
  • Provider choice is per framework, not one unified gateway. The available protocol depends on the selected agent backend.
  • Specialist roles are personal-only. There is no shared public commons or cross-machine specialist forking yet.
  • The task SDK is a first-generation surface. Task creation, polling, and artifact retrieval work; broader orchestration remains open.
  • Switching agent backends cannot transfer in-flight tool state. Existing conversation history can replay, but a running action is not migrated.
  • Skills are local only. There is no shared public commons, cross-machine forking, or user-facing version pinning yet.
  • The reviewer is opt-in and record-scoped. It does not replace domain-specific validation of citations, units, statistics, or methods.
  • Scoped permissions cover allow-grants only. Network sandboxing, directory-level file access control, and a credential vault are not built yet.
  • Windows builds are unsigned. SmartScreen may warn on first launch; official macOS builds are notarized.
  • No local GPU compute backend.
  • No multi-user real-time collaboration.

🙏 Acknowledgements

Thanks to @ewen-poch, @roxi3906, @LJLJ111, @EuanTop, @daanveer-tech, and everyone in Discord, X, and Discussions.

What's Changed

Full Changelog: https://github.com/aipoch/open-science/compare/v0.8.1...v0.9.0

What's Changed

New Contributors

Full Changelog: https://github.com/aipoch/open-science/compare/v0.8.1...v0.9.0

Source: README.md, updated 2026-07-31