An open-source, self-hosted Fathom alternative

Backchannel is an MIT-licensed AI meeting assistant you run yourself with Docker Compose, with desktop builds for Windows, macOS, and Linux. It captures your microphone and, optionally, tab or system audio directly in the browser -- no bot joins the meeting, on any platform, so it works with Zoom, Google Meet, Teams, or anything else. Voice activity detection and speaker diarization run locally on every call, and transcription can run fully offline with local ONNX Whisper or Parakeet models.

Fathom is a cloud service, and its privacy policy is candid about what that means. Its services are hosted in the United States, and de-identified meeting content may be used to train Fathom's in-house models depending on your account settings. To Fathom's credit, the same policy states plainly that it does not authorize third parties such as OpenAI, Anthropic, or Google to train their models on your meeting content. But the structural fact remains: your conversations are on someone else's infrastructure, under terms they set and can revise.

With Backchannel there is no vendor in that position. Audio is captured in your browser, diarized on your server, and stored on your disk. With a local transcription model, even the speech-to-text runs offline. The only outbound calls are the ones you configure to the model provider whose key you supply.

Live summaries versus live advice

Fathom's 2026 desktop update added live summaries that update as the call unfolds, plus a scratchpad for your own notes. That is real and it is useful. But it is a better record of what already happened. Backchannel's agents are pointed the other way: a consolidated analyst on a 40-second cycle proposing questions and observations, an objection handler running every 10 seconds over just the last 90 seconds of conversation and pairing an immediate suggested response with the underlying concern, an opportunity specialist matching what you heard against your own offerings catalog, a synthesizer reconciling insights and noticing when a question has been answered, and a strategic signals agent producing risk and next-question cards every 45 seconds. Fathom's own roadmap lists real-time coaching as coming soon, which suggests they agree this is where the category is going.

Bot-free is no longer the differentiator

We should be straight about this, because a lot of comparison pages are not. Fathom added bot-free capture and it works well. The remaining gap is narrow and platform-shaped: as of July 2026 it is Mac only, with Windows still listed as coming soon, whereas Backchannel's browser capture has never involved a bot on any operating system. If you are on a Mac and the bot was your only complaint, Fathom already solved your problem.

What you give up

Honestly: quite a lot. Fathom is used at 300,000-plus companies and has SOC 2 Type II, GDPR, and HIPAA compliance with SSO and SCIM -- badges a procurement team understands and Backchannel simply does not have. Fathom does 38 languages; we market no language matrix at all. Fathom syncs to HubSpot, Salesforce, and Close, has clips, playlists, and search across every call, and integrates with ChatGPT and Claude. Backchannel offers exports, a cross-session chat, and a REST API. Fathom installs in a minute; Backchannel is a Docker Compose stack, and our prebuilt desktop downloads need an approved Backchannel account rather than being a direct download.

One rough edge worth naming

Backchannel's analysis agents can now run against a self-hosted OpenAI-compatible server (Ollama, LM Studio, vLLM), so with local ONNX transcription the whole pipeline works with the network off and no key from anyone. The wrinkle is that our Privacy First switch does not yet know that: it recognizes only the local transcription models, so switching it on still disables every outside call, including the agents. If you need summarization with zero cloud calls, look at the other open-source options -- some of them do that better than we do.

Setup is a git clone and docker compose up --build, plus a Gemini or OpenAI API key (or a local ONNX Whisper/Parakeet model for fully offline transcription). The quickstart is honest about the friction: first start builds images and downloads models, so give it a few minutes, not seconds. Compared with installing Fathom, this is the harder path and we are not going to dress that up.

Past meetings: Backchannel imports transcripts (.txt, .md, .docx) and audio files (.mp3, .m4a, .wav, .ogg, .flac), and can run its analysis over imported content. Because raw audio is retained, a session can also be re-transcribed later through a different model. There is no one-click Fathom migration tool -- you export from Fathom and import what you care about.