backchannel
Release archive

Every Backchannel release.

Version history for the open-source, self-hosted AI meeting assistant: desktop downloads, publish dates, and what changed in each release, newest first.

Release history

Each release page carries its own downloads, changelog, and first-run notes.

VersionPublishedHighlights
v0.6.5September 4, 2026Long audio imports and Re-transcribe jobs now show progress and can be canceled without replacing an existing transcript with a partial result. End Call completes with transcript refinement enabled, safe Ask requests get one local PII-token repair before the tripwire checks again, connection and update states use plain language, and unmatched voices stay unknown instead of being mislabeled.
v0.6.4September 4, 2026Both exports rebuilt around the reader: the insights workbook opens on a summary sheet instead of an eighteen-column grid, and the call briefing became a document that leads with what came out of the meeting and keeps the transcript line behind every claim. Sidebar groups stay collapsed, the Privacy tab stopped reporting transcript refinement as a gap it never was, and a meeting costs noticeably fewer tokens to analyse.
v0.6.3September 3, 2026Local transcription produced no transcript at all in v0.6.1 and v0.6.2: the bundle was missing files the speech engine reads on every job, and the app reported itself ready anyway. Recordings from those meetings are recoverable by re-transcribing, which no longer erases a transcript it cannot replace.
v0.6.2September 3, 2026Creating a session could stick on "Creating..." forever behind a background model download that deadlocked; saving text no longer waits on a download at all. Model downloads are visible now, with progress, a reason when one fails and a Retry button, and local transcription works in the desktop app again.
v0.6.1September 3, 2026In-app updating no longer asks for an account: pressing Download update sent you to a sign-in for a build the portal hands to anyone, and a v0.6.0 install can now fetch this release on its own. Post-call Chat searches meetings by date like the sidebar, and the Privacy First review panel is readable in dark mode.
v0.6.0September 2, 2026The PII Shield: names, companies, contact details and identifiers become tokens before any model, local or cloud, or the database sees them, with on-device detection, an encrypted vault, honest audio coverage and an audit trail; session setup puts the action button at the top, and the sidebar search understands dates.
v0.5.4September 2, 2026A completed session opens on an executive Overview whose counts link to the rows behind them, the session sidebar is rebuilt with search and a quiet row menu, the briefing reads as one sheet, live-gateway token usage is counted per turn with audio and cached tokens priced at their own rates, and the local API refuses other websites and unknown hostnames.
v0.5.3August 25, 2026The quieter live call screen, built after the v0.5.2 bundles were cut, ships for the first time: three strategic signals ranked by the model instead of five in a fixed order, every signal filed as an insight you can star, dismiss and export, and a transcript column that folds to a rail.
v0.5.2August 16, 2026Supersedes the undistributed v0.5.1 and carries all of it: cost reporting in the Tokens tab including the per-minute live audio bridge, duplicate transcript entries suppressed on dual-track calls, and bounded analysis replies. Adds a post-call briefing whose item headings get the width of their card.
v0.5.0August 2, 2026First-run setup asks which provider you use instead of seeding one, strategic signals persist live and post-call, the briefing names speakers instead of identifiers, and every build path installs the dependency tree that was tested.
v0.4.0July 28, 2026Updates itself with a verified signature and rollback, measures whether the machine can keep up before a call, and carries Privacy First across agents, Analyze, Enhance Insights, and the briefing.
v0.3.8July 25, 2026Self-hosted models become first-class: every endpoint model shows by name, register more than one endpoint, and Privacy First keeps on-prem agents running.
v0.3.7July 25, 2026Analysis agents can run against a self-hosted OpenAI-compatible server such as Ollama or LM Studio.
v0.3.6July 23, 2026Completes OpenAI support across every agent and adds OpenAI batch transcription options.
v0.3.5July 23, 2026Actionable provider failures, refreshed OpenAI model lineup including GPT-5.6, and cost estimates for completed calls.
v0.3.4July 23, 2026Hotfix that repairs starting a call, broken in v0.3.1 through v0.3.3.
v0.3.3July 23, 2026Live Strategic Signals split from the post-call briefing trio, complete agent controls, diarization diagnostics repairs.
v0.3.2July 23, 2026Bundles ffmpeg with the Windows and Linux desktop builds; clearer audio-conversion errors.
v0.3.1July 23, 2026Guided first-run provider setup, default-browser app window on Windows, safely deletable session groups.
v0.3.0July 23, 2026Local voice enrollment for mic-only speaker identity, split End Call flow with on-demand briefings, per-call token usage.
v0.2.5July 22, 2026Dedicated desktop-app window and the latest Gemini Flash model defaults.
v0.2.4July 15, 2026Improved two-speaker diarization, key-free local ASR first use, lighter runtime memory.
v0.2.3July 13, 2026Stable live audio startup and split-track speaker attribution for mic-plus-system-audio calls.
v0.2.2July 13, 2026Desktop delivery moved to the authenticated Backchannel portal with stronger installer access controls.
v0.2.1July 11, 2026Brand icon on the desktop apps, session data protected across upgrades, automated Linux tarball release.
v0.2.0July 11, 2026Private early-access administration, richer meeting chat context, portable Linux x64 desktop bundle.
v0.1.1July 11, 2026Better call continuity, chat readability, speaker attribution, interface contrast, and Windows launch experience.
v0.1.0July 10, 2026First desktop release: Windows x64 and macOS Apple Silicon executables without Docker.

Run the latest release on your own hardware

Desktop bundles for the fastest start, or check out the matching source tag and rebuild Docker Compose for maximum control.