Backchannel vs Anarlog (formerly Hyprnote)

Privacy is the price of entry here, not the differentiator

Anarlog transcribes on-device and keeps recordings on your machine. So does Backchannel. So do Meetily, Buzz, and Vibe. Anyone comparing local-first meeting tools should assume every serious option in the category already clears that bar, and then ask a harder question: once the audio is captured locally, what does the tool actually do with it? Anarlog turns it into a good record. Backchannel turns it into help while the meeting is still running.

Who said what, and who charges for it

Anarlog's documented free workflow for speaker labels is manual: you pick a speaker label in the transcript, assign a participant, and apply it to their other turns. Automatic speaker indexes arrive from whichever transcription provider you configure, and hosted speaker identification sits in Anarlog Pro. A pull request adding local pyannote diarization to the repo was closed without merging on 2026-05-01. Meetily made the same call from the other direction: its diarization shipped, but only in the paid Pro edition. Two of the best-known open-source meeting projects have now routed automatic speaker attribution to a paid or hosted tier.

Backchannel gives it away. Silero VAD segments speech and WeSpeaker embeddings assign each segment to a voice, live, in the MIT codebase, with dual-track capture giving you and remote participants separate identities and an optional enrolled voice profile for your own. To be clear about what is and is not special here: the underlying stack is not exotic -- the open-source Vibe project uses a comparable pyannote-plus-WeSpeaker approach for files. What is unusual is running it live, during the call, on split tracks, for free, and feeding the result straight into analysis.

Notes after vs insights during

Anarlog's center of gravity is the record. You type rough notes in a memo pane while it listens, and afterwards it merges those notes with the transcript into a structured summary you can chat with. It is a genuinely good workflow and people who live in markdown like it a lot.

Backchannel keeps the record too, but its center of gravity is the call itself. Nine seeded agents ship in the box: an audio gateway for interim transcription, five live insight agents, and a three-agent briefing pass that runs when the call ends. The live five are the consolidated analyst (every 40 seconds), the objection handler (every 10 seconds, over only the last 90 seconds of transcript), the synthesizer, the opportunity specialist, and strategic signals (every 45 seconds). Each has its own model, prompt, and trigger, editable per deployment and per session. No actively maintained open-source meeting tool ships this today. Amurex (AGPL-3.0) did, and it has had no commits since May 2025.

What Anarlog means by "agents"

Worth stating precisely, because the word collides. Anarlog ships a local MCP server and a CLI so that external agents -- a coding assistant, a script, your own tooling -- can list recent meetings, search by title, and read transcript pages in bounded chunks. The documentation is explicit that all current Anarlog data operations are read-only. It is a well-designed integration surface for querying stored meetings after the fact, and if that is what you need, it is arguably cleaner than anything Backchannel offers. It is not in-call assistance, and Anarlog does not claim it is.

macOS only, and openly so

Anarlog's shipping artifacts are macOS disk images, for both Apple Silicon and Intel. Its May 2026 rename post states plainly that Linux and Windows support are not planned. That is a legitimate product decision -- focus is why the Mac app is as good as it is -- but it is also the clearest fork in the road between these two projects. Backchannel ships Windows, macOS, and Linux desktop bundles and a Docker Compose stack that runs anywhere. If your team is not all on Macs, this decides it before any feature does.

Desktop app vs server

Anarlog is an appliance: one machine, one user, local SQLite, nothing to operate. Backchannel is infrastructure: a FastAPI backend and PostgreSQL behind a browser UI, so one deployment can serve a team, keep a shared archive, expose a REST API, and answer questions across up to twenty past sessions at once. Neither is better in the abstract. Pick by how many people need it and whether you want something to run.

The offline story

Both sides can now do this. With an on-device transcription model plus LM Studio or Ollama, Anarlog records, transcribes, summarizes, and chats with no internet connection at all. Backchannel's diarization is always local, transcription runs offline with local ONNX models, and the insight agents can point at any OpenAI-compatible server -- Ollama, LM Studio, vLLM -- so the live analysis runs on your hardware too, with no key from anyone. One honest wrinkle on our side: the Privacy First switch still turns the agents off, because it recognizes only the local transcription models. A fully local Backchannel is configured through the endpoint setting rather than by flipping that switch.

Maturity and community

Anarlog wins, clearly. About 8.9k GitHub stars, a very low open issue count, and releases shipping nearly every day. It has also been through two renames in 2026 -- Hyprnote to Char to Anarlog -- which is worth knowing when you go looking for documentation or old forum threads, and it relicensed from GPL to MIT along the way. Backchannel is far younger and far smaller. What it offers is capability that is not in Anarlog at all; weigh that against community support according to your own risk tolerance.

Anarlog: download the macOS disk image, open it, download an on-device model. Two steps, no terminal, no account, no key. Backchannel: git clone, docker compose up --build, add a Gemini API key (or a local ONNX transcription model), open localhost in a browser -- or use a desktop bundle for Windows, macOS, or Linux, which needs an approved Backchannel account where Anarlog's download is ungated. Call it five to seven steps against two, and the quickstart is honest that the first start builds images and downloads models. Anarlog's install experience is better and it is not close. Backchannel's payoff is a shared, multi-user, cross-platform server once it is up.