Skip to content

OmniBioAI Studio v0.9.0-beta

Pre-release
Pre-release

Choose a tag to compare

@man4ish man4ish released this 16 Nov 07:01
e779034

OmniBioAI Studio v0.9.0-beta

OmniBioAI Studio v0.9.0-beta is an evaluation release. It moves literature search to the PubMed RAG v6 indexes, fixes plugin correctness issues found by the plugin testkit, and continues the platform hardening work since v0.8.2-beta.

📦 Downloads

Platform Architecture File
Linux x64 OmniBioAI.Studio-0.9.0.AppImage
Linux x64 omnibioai-studio_0.9.0_amd64.deb
Linux x64 omnibioai-studio-0.9.0.x86_64.rpm
Linux ARM64 OmniBioAI.Studio-0.9.0-arm64.AppImage
Linux ARM64 omnibioai-studio_0.9.0_arm64.deb
Linux ARM64 omnibioai-studio-0.9.0.aarch64.rpm
macOS Intel OmniBioAI.Studio-0.9.0.dmg
macOS Apple Silicon OmniBioAI.Studio-0.9.0-arm64.dmg
Windows x64 OmniBioAI.Studio.Setup.0.9.0.exe

All nine installers are built by GitHub Actions from the v0.9.0-beta tag. Check every download against SHA256SUMS-v0.9.0-beta.txt. The macOS and Windows installers are unsigned and not notarized.

🚀 Quick Start

Linux (DEB - Ubuntu/Debian):

sudo dpkg -i omnibioai-studio_0.9.0_*.deb
omnibioai-studio

Linux (RPM - RHEL/Fedora):

sudo rpm -i omnibioai-studio-0.9.0.*.rpm
omnibioai-studio

Linux (AppImage):

chmod +x OmniBioAI.Studio-0.9.0*.AppImage
./OmniBioAI.Studio-0.9.0*.AppImage --no-sandbox

macOS: open the DMG, drag the app to Applications, then right-click → Open (the app is not notarized).

Windows: run the installer and launch from the Start menu (the installer is unsigned).

📋 Requirements

  • Docker (auto-prompted on first launch)
  • 16GB RAM minimum (32GB recommended)
  • 50GB free disk space
  • License key (get one at omnibioai.org)

🔑 Get Beta Access

Sign up at: https://omnibioai.org/#request

Highlights

Literature search on PubMed RAG v6

  • New per-study PubMed search, using Qwen3-Embedding-4B indexes, one per study. Results are keyed by PMID and include the title, year, journal, a snippet and the studies each hit came from (omnibioai-rag#48).
  • The Lit AI page now uses the v6 catalog and search. It has a studies sidebar with filter and multi-select (up to 8 studies), result cards, and Index Stats showing the model, dimension, vector counts and docstore status (omnibioai-rag#51).
  • The API Gateway literature routes (/v1/literature/search, /studies, /domains) now use RAG v6. Public response shapes are unchanged. Snippets are now real excerpts instead of full abstracts (omnibioai-api-gateway#37).
  • The Bio Agent's literature mode calls RAG v6 over HTTP (omnibioai-workbench#926).
  • Literature search requires a signed-in user with dataset read permission, and is rate-limited per user. Logs record the query length, not the query text.

Plugin fixes

These fixes come from the plugin testkit runs. Each fixed plugin now has a regression test that runs on its own testkit data.

  • Plugins now report a failure when their input is missing or a required container step fails. Before, some returned synthetic or placeholder results marked as successful (single-cell import and trajectory, variant annotation, metabolomics stats/normalize/imputation/outlier detection) (omnibioai-workbench#927, #931).
  • Run manifests for CyTOF analysis, HRD score and ctDNA liquid biopsy now pass schema validation, so correct runs are no longer marked failed (#927).
  • Statistical corrections:
    • Metabolomics outlier detection uses a robust PCA-based test that works when metabolites outnumber samples.
    • Imputation uses each metabolite's own minimum.
    • Benjamini–Hochberg adjustment is applied correctly in ATAC differential, motif enrichment, ATAC report and DMR analysis.
    • Proteomics QC CV excludes missing values.
    • ATAC peaks no longer cross chromosome boundaries.
    • DMR delta-beta follows the declared group order (#931).
  • Matching corrections:
    • Clinical trial matching uses whole-word terms.
    • Oncoprint classifies variant consequence terms exactly.
    • HRD LST counting requires a change in copy-number state (#927).

Security hardening

  • Several plugin and integration APIs are temporarily restricted at the router until per-user authentication is added. The format converter and workflow builder are read-only; the Benchling integration API is closed; the ServiceNow integration API is read-only; the genome viewer track API is unavailable (omnibioai-studio#282, #290, #292).
  • The Bio Agent no longer writes query text to logs (omnibioai-workbench#923, fixed in #926).
  • Deployment tooling keeps secret values out of terminal output and logs (omnibioai-ecosystem#111).
  • Search indexes and shared reference data are now mounted read-only into the services that read them.
  • Platform signing keys were rotated on 2026-11-16 and the rotation was verified: tokens signed with the previous key are rejected on every endpoint, and a fresh sign-in and Bio Agent search work with the new key. All sessions from before the rotation ended, so every user signs in again once.

Behaviour changes

  • A study is required for literature search. POST /v1/literature/search returns 400 invalid_request (field domain) when no study is given. There is no "all studies" search in v6. Unknown study names return 400 and are not billed. Legacy names such as "DNA damage response" still resolve.
  • Retired legacy literature features:
    • The legacy embedding and knowledge-graph build paths are disabled.
    • On the Lit AI page, ingest (Ingest study, Fetch abstracts, New study), Embed, Benchmark and RAG+LLM answers are disabled, along with the features that depend on the retired legacy indexes.
  • /v1/literature/answers is unchanged, but RAG v6 does not generate answers yet.
  • When the per-user literature search limit is hit, the API returns 429 rate_limited with Retry-After.

Known issues

Read the attached KNOWN_ISSUES_V0.9.0_BETA.md before installing. This prerelease is intended for evaluation and does not claim production, regulatory, or HIPAA qualification.

Full Changelog: v0.8.2-beta...v0.9.0-beta