v0.3.1 — fix alphafold.fetch (signed + notarized)

OpenScience v0.3.1 — bugfix

Signed and notarized macOS build (Developer ID). MIT licensed. No paid API keys required.

Fixed

• alphafold.fetch returned HTTP 404 when downloading the structure file. It was constructing file URLs at /api/entry-pdb/{acc}, which no longer exists. It now reads the authoritative pdbUrl/cifUrl from the AlphaFold DB metadata endpoint (/api/prediction/{acc}) — correct file-name case (AF-{acc}-F1-model_v{N}) and current model version — and downloads via that.
• Missing predictions (e.g. Q8WZ42 and other very large proteins that aren't in the AlphaFold DB) previously surfaced as a raw HTTP 404. They now return a clear, actionable error suggesting pdb.fetch for an experimental structure.
• The result now also includes description, organism, and modelEntityId.

Verify

• Live-checked: P12345 downloads and renders in the 3D viewer (mean pLDDT 94.12); Q8WZ42 returns the clear "no prediction" error.
• 44 sidecar tests + 4 frontend tests pass.
• Signed build passes spctl --assess (source=Notarized Developer ID) and xcrun stapler validate.

Install (macOS Apple Silicon)

Download OpenScience_0.3.1_aarch64.app.zip, unzip, drag OpenScience.app to Applications, open. Signed + notarized — no Gatekeeper workaround needed. First launch takes ~10–20s while the bundled Python runtime starts (status pill shows "starting…").

openscience-0.3.1-source.tar.gz is the full source. See Signing the app in the README to build/sign yourself.

v0.3.0 — Session branching, agents, skills, approval gate, provenance (signed + notarized)

OpenScience v0.3.0

Signed and notarized macOS build (Developer ID). MIT licensed. No paid API keys required.

New (second feature list — no user keys/infra needed)

• Session branching — fork a run into a new branch from the Run Inspector (like git branches); the fork records parent_run_id and seeds its conversation from the parent.
• User-created specialist agents — define an agent with a system prompt and a tool whitelist in Settings; activate it from the top bar.
• Reusable skills — save a prompt/pipeline as a skill and re-invoke it from the top bar.
• Approval before spending compute — when enabled in Settings, compute.run/slurm.submit draft a plan and wait for confirmation before running jobs or submitting to HPC.
• Plain-English provenance — every code.run figure records the source code, tool, backend, and how it was generated; viewable in the Run Inspector audit trail.

Fixes

• Signed sidecar crashed on launch ("different Team IDs") — the hardened runtime refused to load the bundled libpython. Fixed by signing the PyInstaller sidecar with an entitlements plist (disable-library-validation, allow-unsigned-executable-memory). The signed app now boots to 25 tools.
• "sidecar: offline" on launch — the UI now shows "sidecar: starting…" (amber) during the ~10–20s Python runtime cold start, then "online". Port discovery self-heals when port 7100 is held by a stale process.
• UI polish: top-bar agent/skill selectors styled as pills, scrollable Settings modal, gradient/spacing refinements.

Install (macOS Apple Silicon)

Download OpenScience_0.3.0_aarch64.app.zip, unzip, drag OpenScience.app to Applications, and open. It is signed and notarized — no Gatekeeper workaround needed. The app bundles its own Python sidecar (no Python/uv required). First launch takes ~10–20s while the Python runtime extracts; the status pill shows "starting…" during this.

openscience-0.3.0-source.tar.gz is the full source. To build/sign yourself, see Signing the app in the README.

Verify

43 sidecar tests + 4 frontend tests pass. Tool dispatch + viewer SSE verified end-to-end. Signed build passes spctl --assess (source=Notarized Developer ID) and xcrun stapler validate.

v0.2.0 — Code execution, compute, databases, literature, citations, manuscripts

OpenScience v0.2.0

Open-source scientific AI workbench. MIT licensed. No paid API keys required — every data source is a free public API and code runs locally (pair with Ollama for a fully free/local stack).

What's new (18 new tools, 25 total)

• Code execution & figures — code.run executes Python/R on your machine or over SSH/Slurm, captures output, and renders matplotlib/plotly figures in a new FigureViewer. Editable-by-chat: ask "make the x-axis log scale" and it rewrites the plotting code.
• Compute management — compute.run + slurm.submit/status/cancel wire the existing local/SSH/Slurm backends to the agent. Remote code.run stages scripts over SSH and pulls figures back via SFTP.
• More databases — Ensembl, ClinVar, GEO, AlphaFold DB (reuses the 3D protein viewer).
• Literature — PubMed + Europe PMC search with abstracts.
• Citations — Crossref metadata + BibTeX/RIS/CSL export for properly cited reports.
• Manuscripts — pin chat drafts into sections, insert figures and citations, export to Markdown/LaTeX/PDF (via pandoc, with graceful fallback). Assembled manuscript + bibliography are saved to the run for reproducibility.

Fixes

• Robust sidecar port discovery (IPC + scan fallback) — the UI now self-heals when port 7100 is held by a stale process (previously showed "sidecar offline" and no run history).

Install (macOS Apple Silicon)

Download OpenScience_0.2.0_aarch64.app.zip, unzip, drag OpenScience.app to Applications. The app bundles its own Python sidecar (no Python/uv needed). It is unsigned — on first launch, control-click the app and choose Open, then confirm in the Gatekeeper dialog. See the signing notes below for producing a signed build.

openscience-0.2.0-source.tar.gz is the full source for this release.

Verify

37 sidecar tests + 4 frontend tests pass. Tool dispatch + viewer SSE verified end-to-end against a live PDB fetch.

OpenScience v0.1.1 (macOS Apple Silicon)

macOS Apple Silicon release. API keys are now stored in macOS Keychain, RDKit chemistry assets are bundled for offline use, PDB/UniProt/NCBI/ChEMBL HTTPS requests use the macOS system trust store, and runs can verify output hashes. Includes a self-contained sidecar: no Python or uv is required. Unsigned build: control-click > Open on first launch.

OpenScience v0.1.0

First macOS desktop release for Apple Silicon. Includes a self-contained native Python sidecar, so no Python or uv installation is required. Download the DMG or the .app ZIP. This unsigned build may require right-click > Open on first launch.