# anarlog: a local-first, bot-free meeting notetaker you can fork > anarlog is the MIT-licensed community app from the fastrepl team, a Granola alternative that captures device audio instead of sending a bot into the call. Here is what it does, how to build it, and where it stops. **fastrepl/anarlog** — Open source Granola AI Alternative. How to use it Download the latest release for your platform: github.com/fastrepl/anarlog/releases/latest Open it and join a meeting. - Repository: https://github.com/fastrepl/anarlog - Website: https://anarlog.so - Stars: 9,395 · Forks: 771 - Language: TypeScript - License: MIT - Published: 2026-08-04 · Updated: 2026-08-18 · Language: en - Canonical page: https://hysenlabs.com/projects/fastrepl-anarlog ## The problem anarlog targets: a notetaker that never enters the room Most AI meeting tools work by sending a participant into the call. That participant appears in the roster, records from inside the meeting, and usually ships the audio to a vendor. anarlog takes the opposite route. The README states that it "takes notes in your meetings without sending a bot to join the call: it listens to your device audio, can transcribe on your machine, and keeps app data in a local SQLite database." There is no participant to explain to the other side of the call. The intended user is explicit in the README: people who want AI meeting notes without handing conversations to someone else's cloud, and anyone who has to get a notetaker past a security review. That second group is the more interesting one. A tool that captures on the device and stores to local SQLite is a different conversation with an infosec reviewer than a SaaS recorder. anarlog is not trying to win on features against hosted notetakers. It is trying to be defensible. ## How anarlog splits transcription from summarisation The architecture separates two stages that many competitors bundle. Stage one turns audio into a transcript; stage two turns transcript plus your memo into a summary, a title, or chat answers. The README is direct about this: "anarlog keeps audio transcription separate from the language model used for summaries and chat. You can change either one without changing the other." Transcription has three routes. Managed Pro (Cloud) picks a provider by language and by live or batch mode, and the README names Deepgram Nova, Soniox 5, and AssemblyAI Universal 3.5 as current primary paths. On supported Macs, Soniqo or Apple Speech can run on-device. You can also select your own transcription provider and model. Summarisation has two routes. Managed Auto, also shown as Pro (Cloud), currently uses the latest Claude Sonnet alias through OpenRouter. Otherwise you supply an API, a subscription, an OpenAI-compatible server such as Ollama, LM Studio, or Unsloth, or an eligible Apple Intelligence model. The active provider and model are always visible under Settings, in the Transcription and Intelligence panes. That split is the design decision worth caring about. Swapping the summariser is cheap and reversible. Swapping the transcriber changes what leaves your machine, so the two settings carry very different privacy weight even though the UI presents them side by side. The repository map matches the README's claims: apps/desktop is a Tauri v2 app with a React and TypeScript UI over a Rust backend, crates/* holds Rust libraries for audio capture, transcription, diarization, storage and services, and plugins/* holds Tauri capabilities such as local STT, database access, calendar, export and notifications. ## Installing anarlog and running a first meeting For normal use you do not build anything. The README's get-started path is to download a release for macOS, Windows, or Linux from anarlog.so/download, or from the latest GitHub release. Open it, join a meeting, and anarlog records on your device and transcribes with the model you selected. You then generate a note, edit it like a document, and export Markdown. Before that first meeting, open Settings and check the two panes the README calls out. Settings → Transcription shows the active transcription provider and model. Settings → Intelligence shows the active summary and chat provider. Both are visible at all times, which is the point: you should know whether audio is being transcribed locally or by a hosted provider before you record anything sensitive. Building from source is a different job. The README requires Node.js 22 or later, pnpm 11.1.1, Rust 1.94.0, and the Tauri v2 system dependencies. On Debian or Ubuntu the repository ships a script that installs the toolchains and system packages: ```bash bash scripts/setup-linux.sh ``` After that, install the workspace and start the desktop app: ```bash pnpm install --frozen-lockfile pnpm exec turbo dev:desktop ``` The package.json confirms the toolchain expectations: packageManager is pnpm@11.1.1 and engines.node is >=22. The dev:desktop script resolves to pnpm -F @anlg/desktop tauri:dev. If you want the website instead, the README gives pnpm exec turbo dev:web. The README notes that the local-first desktop app and website start without secrets, while hosted AI, CloudSync, authentication, billing and connected integrations need optional local services and configuration. So a fork that only needs on-device capture and a local model does not require you to stand up Supabase or the hosted API. Read CONTRIBUTING.md for the validation commands and ownership rules before opening a pull request. ## Where anarlog is the wrong tool The bot-free design has a cost. Capturing device audio means the app records what your machine plays and what your microphone hears. On a laptop that is fine. In a conference room with a shared speakerphone, or on a call where you join from a phone and the laptop is closed, there is nothing for anarlog to capture. A bot that dials in does not have that problem. If your meetings do not happen on the machine running anarlog, this is the wrong category of tool. The platform story is also uneven in the README. It says on-device transcription models are available "on supported Macs," and lists Soniqo or Apple Speech plus eligible Apple Intelligence as local options. Windows and Linux are listed as download platforms, but the README does not make the same local-model claim for them, so the privacy-first configuration may be narrower than the download page suggests. Verify per platform rather than assuming parity. The repository carries apps/mobile and apps/watch/apple as source, and the README is explicit: "Mobile client source; no mobile app is currently distributed." If you need a phone notetaker today, that source tree is not a product. Finally, read the licence boundary before you plan around it. The community application is MIT, but the README states that source-visible enterprise components are commercially licensed, and the repository has both LICENSE and LICENSE.enterprise at the top level alongside LICENSING.md. The README's own framing about giving security and IT teams "a real path to yes" describes a commercial motion sitting next to the open source one. ## anarlog against Granola and Meetily Granola is the comparison the project invites. The README describes anarlog as "an open-source alternative to Granola" and the tagline reads "Granola, rearranged." The practical difference is control of the pipeline. With Granola you get a managed product and a managed model choice. With anarlog you get the same shape of workflow (join a meeting, get a note, edit it, export) but you decide which transcription provider runs and which LLM writes the summary, and you can point the Intelligence setting at an OpenAI-compatible server on your own hardware. The trade is that you also own the configuration, the API keys, and the failure modes. Meetily is the other name that comes up in searches around this project, and it is a different bet. anarlog is a desktop application built on Tauri with a Rust core, distributed as platform installers, with a CLI and MCP server in apps/cli for agent access. If your reason for choosing an open source notetaker is that you want to run it as a service, Meetily's Docker-oriented packaging is the axis to compare, because anarlog's README documents a desktop app, a website, and optional hosted services, not a container image. Compare on deployment shape first, then on transcription and summary providers. One more alternative worth naming: doing nothing and keeping raw recordings. anarlog's export is Markdown, and sessions, notes and transcripts live in local SQLite with recordings and attachments as plain local files. If you already have a transcription pipeline you trust, the value anarlog adds is the capture layer and the note editor, not the storage format. ## Maintenance, upgrades and licence cost The last push to the repository was on 2026-08-29, and the most recent release in the release list is desktop_v1.4.15 on the same date, following desktop_v1.4.14 on 2026-08-28 and desktop_v1.4.13 on 2026-08-24. That is a recent cadence with closely spaced desktop builds, and the repository is not archived. What the README does not document is rollback. There is no stated procedure for reverting a desktop release, and no migration notes for the local SQLite database across versions. If you run anarlog on a work machine with a long note history, that gap matters more than the release frequency. Back up the SQLite file and the recordings directory yourself before upgrading, because the README does not promise that an upgrade is reversible. The README also records a rename history: the project started as Hyprnote, briefly used the char name, and the team is now building char as a separate productivity app while anarlog continues as the open source local-first notetaker in this repository. That split explains the enterprise directory and the commercial licence file. For an adopter, the practical question is which parts of your workflow sit inside the MIT grant. The community application is MIT-licensed and the README says you can fork it, audit it, sell it, or build it yourself, but LICENSING.md and LICENSE.enterprise exist precisely because not everything in the tree is under the same terms. Read them before you vendor anything from enterprise/. ## Conclusion anarlog fits engineers and privacy-conscious teams who want meeting notes without a bot in the participant list, and who are willing to pick their own transcription and LLM providers. Skip it if you need the polished hosted experience of a commercial notetaker, if you expect a supported mobile app (the repository says no mobile app is currently distributed), or if you need a formal support contract, since the README points enterprise components at a separate commercial licence. Before adopting it, check the release page for a build for your platform, confirm in Settings which transcription and Intelligence providers are active on your machine, and read LICENSING.md to see where the MIT grant ends. ## FAQ ### What are some open source meeting transcription software options? anarlog is one: the README describes it as an open-source, MIT-licensed alternative to Granola that captures device audio rather than sending a bot into the call, with local SQLite storage and Markdown export. The README also lists Ollama, LM Studio and Unsloth as OpenAI-compatible local servers you can point the Intelligence setting at. ### Does anarlog send a bot to join my meeting? No. The README states that anarlog captures audio directly on your device, so nothing appears in the participant list and nothing records from inside the meeting. The trade-off is that the meeting has to be audible on the machine running anarlog. ### Can anarlog transcribe meetings entirely on my machine? The README says that on supported Macs, available built-in transcription models run on-device, naming Soniqo or Apple Speech, and that Local Intelligence providers keep summaries and chat on your computer too. The active provider and model are always visible under Settings → Transcription and Settings → Intelligence. ### How do I install anarlog? The README points to the download page for macOS, Windows, and Linux, and to the latest GitHub release. Open the app and join a meeting; anarlog records on your device and transcribes with the model you selected. Building from source instead requires Node.js 22 or later, pnpm 11.1.1, Rust 1.94.0, and the Tauri v2 system dependencies. ### Is anarlog free to use and fork? The README says the community application is MIT-licensed and that you can fork it, audit it, sell it, or build it yourself. It also states that source-visible enterprise components are commercially licensed, and the repository carries LICENSE.enterprise and LICENSING.md alongside LICENSE, so check which parts you are using. ## Sources - [Official documentation](https://anarlog.so) - [Official README](https://github.com/fastrepl/anarlog#readme) - [Project repository](https://github.com/fastrepl/anarlog) - [Release notes](https://github.com/fastrepl/anarlog/releases) --- Hysen Labs editorial analysis, written from the project's own repository and release notes. Cite the canonical page: https://hysenlabs.com/projects/fastrepl-anarlog