Download the Liminon beta.

What's new in 1.18.1

• Cleaner transcript segments — transcripts now break at sentences and speaker changes from the very start of a recording (previously the first ~30 seconds could mix both speakers in one segment). Re-transcribe a recording to apply it.
• No more duplicated text in long recordings — fixed repeated text where recordings longer than 3 minutes are stitched together.
• Open projects from Finder — double-click a .qproj file in Finder to open it in Liminon.
• Methodology reports name Liminon — "Prepared with" and the citation sentence now read Liminon.
• Real project size in Privacy & Data — the panel now shows the actual size of your project.
• Layout fixes — panes now fill the window on Case Comparison, Findings, Research Questions, Memos and Interview Profiles; Reports & Exports no longer pushes the sidebar off-screen at the default window size; the Case Comparison scope control no longer truncates.

What's new in 1.18.0

• Buy a license from inside the app — the Liminon menu now has "Purchase a License…" (also linked from the License panel). Liminon stays completely free to use during the beta; buying gets you a permanent license key by email to support development and lock in your price — it does not unlock any feature.
• Early-bird / Standard / Student pricing shown in-app — $129 early-bird, $179 standard, $79 student, one-time and tax included. Each Buy button opens checkout for that tier at liminon.ai/pricing in your default browser; no payment details are ever collected inside the app.
• Rotated the license signing key — the previous key pair is retired; future license keys are issued and verified against a new one.
• Update checks and the Privacy panel now point at liminon.ai — was the older sondera.pages.dev deployment domain; same site, no behavior change.

What's new in 1.17.1

• Unified branding — every user-visible name in the app now reads Liminon; a handful of dialogs and onboarding screens that still said "Sondera" have been updated.
• In-app feedback goes straight to us — Send Feedback now opens a message to [email protected].
• Developer ID signed and notarized — same signing and Apple notarization as every beta build.

What's new in 1.17.0 beta

• Unified Reports & Exports center — a single sidebar entry replaces the scattered export buttons; six sections cover transcripts, codebook, matrices, quotations, the methodology report, and one-click full research package.
• Methodology report as Word and PDF — generate a paper-ready methods appendix with a cover page, citation block, numbered pages, and your full audit trail (AI accept/reject ratios, codebook evolution), not just plain Markdown.
• Real Excel workbooks (.xlsx) — matrices ship as a single Excel file with frozen headers, bold totals, code × recording, code × speaker, codebook, and summary sheets, no third-party dependency.
• One-click full research package — zip up every transcript, the codebook, matrices, the quotation table, the methodology report PDF, and the audit CSV into one self-describing archive ready for IRB submission or co-author handoff.
• Ask AI on a selected phrase — when coding a phrase inside a segment, Ask AI now reasons over just that excerpt and writes a range-precise assignment, instead of always evaluating the whole segment.
• Speaker count up front — first-time transcription asks "how many people speak?" so diarization has the hint it needs on real recordings; "Not sure" still auto-detects.
• Video files import natively — drop in an .mp4/.mov and the audio track is transcribed directly, no conversion step.
• Default AI model moved to qwen2.5:3b — materially better code suggestions than the previous 1.5b default; 1.5b still works if it's what you have pulled.
• Plus everything from 1.0 beta: text-selection coding, code trees with merge, ⌘F find-in-transcript, playback follow, semantic search, parallel-chunk 3× transcription, a playable sample project, and undo for structural edits.

Getting started

1. Open the DMG and drag Liminon into Applications, then launch it.
2. Create a project, drop in an interview recording (audio or video), and let the on-device model transcribe and separate speakers. Everything stays on your Mac.
3. Code the transcript by hand, or turn on AI suggestions — see the note on Ollama below.

AI features need Ollama (optional, 5 minutes)

Known limitations in this beta

• English-language interviews work best; other languages transcribe but AI coding quality varies, and speaker separation accuracy on multi-speaker non-English recordings is still being measured.
• AI code suggestions are deliberately conservative — treat them as a starting point for your own judgment; the methodology report tracks every accept/reject for your audit trail.
• REFI-QDA (NVivo/ATLAS.ti) project import is on the roadmap, not in this build.
• Projects are saved as .qproj bundles — back them up like any other research file; there is no cloud sync by design.