Pulse is a voice-first AI side panel for Windows and a markdown notes app for Android. Both run on-device. Both are Apache 2.0. This page is the honest side-by-side: how Pulse stacks up against Raycast AI, Notion AI, AnythingLLM, LM Studio, and bare Ollama. No asterisks, no marketing copy.
Same ten questions, six products, no asterisks. Yellow dots mean partial. Hover on a column header on desktop to see the full product name. The table collapses to cards on phones.
| Feature | Pulse | Raycast AI | Notion AI | AnythingLLM | LM Studio | Ollama (alone) |
|---|---|---|---|---|---|---|
| Platform | Windows + Android | macOS only | Web + iOS + Android | Docker (any OS) | Win / macOS / Linux | CLI / API only |
| Local-first | ||||||
| Voice input | ||||||
| Markdown notes | ||||||
| Price | Free | $8 / mo | $10 / mo | Free | Free | Free |
| Open source | Apache 2.0 | MIT | MIT | MIT | ||
| Mobile | Android | iOS + Android | ||||
| Setup time | ~2 min | ~5 min + sub | ~10 min + account | ~30+ min | ~10 min | ~5+ min |
| RAM for 7B model | ~4-8 GB | n/a (cloud) | n/a (cloud) | ~4-16 GB | ~4-16 GB | ~4-16 GB |
| Privacy | On-device. No tracking. | Cloud. OpenAI terms. | Cloud. Telemetry. | On-device. Self-hosted. | On-device. No tracking. | On-device. No tracking. |
Voice + notes. Local. Apache 2.0.
macOS only. $8/mo. Cloud-tied.
Cloud wiki + AI. $10/mo. Tracking.
Local. MIT. Docker. 30+ min setup.
Local model GUI. MIT. No app layer.
Local model engine. MIT. CLI / API only.
Source: each product's own site and pricing page as of . Prices in USD. "RAM for 7B model" is the typical Q4 / Q5 GGUF run at Q4 quant on a 16 GB machine.
The other tools on this list solve real problems. Pulse is not the answer to every question. Here is what each one is, in one paragraph, without the spin.
Eight dollars a month, macOS only, fast prompt bar. But your prompts go to OpenAI by default. The local-model extension is paid, separate, and still routes through Raycast servers. Good if you live in Raycast and trust the chain.
Ten dollars a month, polished cloud editor with AI baked into every doc. But every note, every prompt, every keystroke is on Notion servers under their terms. The product is the wiki, the AI is the upsell.
Free, MIT, local-first, multi-model. Power-user surface: Docker, vector DB config, workspace permissions. Best choice if you want to roll your own and have a weekend.
Free, MIT, polished local-model GUI. Runs GGUF models on your hardware. Stops at the chat window — no notes, no voice, no app layer. You wire your own front-end on top.
Free, MIT, the model engine everyone else builds on. No UI at all. You talk to it via curl, a script, or a third-party wrapper. Pulse uses Ollama as its local backend on desktop.
Pulse is not a platform. It does one thing well: runs an AI on your device, and keeps your notes on your device. The list below is what Pulse actually does in v0.6.6, not what it might do someday.
Pulse talks to a local Ollama instance on desktop and uses on-device Whisper on Android. No cloud round-trip, no account, no telemetry. The on-device update checker only knows your version number.
Pulse is Apache 2.0 across all three surfaces: Windows desktop, Android notes, and this site. No "Pro" tier, no "Enterprise" tier, no telemetry opt-in, no premium model hidden behind a paywall.
Voice is the desktop side panel. Markdown is the Android editor. Both are offline by default. Both are first-class surfaces, not a feature flag.
There is no analytics SDK. There is no error reporter. There is no "phone home" with usage data. The only network call Pulse makes is the manifest poll for updates, and you can disable it.
This page would be lying if it said Pulse is for everyone. It is not. The honest list of when to pick a different tool is below. No upsell, no "but try it anyway."
You need frontier-scale AI (1M+ token context, latest GPT-4 / Claude). Pulse runs 4-70B local models. Use the cloud tools.
You live in Raycast on macOS and are happy paying $8/mo for the prompt bar. Raycast AI is fine for that workflow.
You do not have at least 8 GB of free RAM for a 7B model. Use a cloud tool or wait for a smaller distilled model.
You need a multi-user wiki with comments, mentions, and shared workspaces. Notion AI is the right tool. Pulse for Android is single-user.
You want a polished desktop chat UI with image generation out of the box. LM Studio plus a chat front-end is a better fit than Pulse.
You want a fully GUI-driven local RAG pipeline. AnythingLLM is the closest match and the config is honest about it.
Yes. Desktop Pulse talks to a local Ollama instance. Android Pulse runs on-device. The only network call is the optional update-checker manifest poll, which you can disable in settings. Voice input on Android uses the on-device speech recognizer; on desktop it uses Whisper.cpp running locally.
Anything Ollama serves on desktop: Llama 3, Qwen 2.5, Mistral, Gemma 2, Phi-3, DeepSeek, and any other GGUF model. The 4-8B range (Qwen 2.5 7B, Llama 3.1 8B, Mistral Nemo 12B) hits the best latency / quality balance on a 16 GB laptop. On Android, Pulse for Android uses its on-device model for the recall step and ships the Smart Engine v3 router.
Not yet in v0.6.x. Pulse for Android v0.7 is the planned E2E sync release, opt-in and disabled by default. Today, the on-device SQLite file is plain and portable — copy notes.db off the device with any file manager, restore the same way. No proprietary export format.
Linux desktop is not in v0.6.x. The desktop binary is Windows (Tauri 2). iOS is not planned for the Notes surface. Android is minSdk 24 (Android 7.0) with targetSdk 36.
No. Pulse is Apache 2.0 across both surfaces. There is no Pro tier, no premium model, no "Enterprise" SKU. The same code is the same code.
There is nothing to host. Pulse is local-first. The desktop app talks to your local Ollama; the Android app runs on-device. The site is a static Astro build. The only network endpoint is the manifest poll for updates, and the chain of hosts is public.
Privacy, cost, and offline use. ChatGPT and Claude route every prompt through their cloud, log it, and charge per token. Pulse runs the model on your hardware, costs nothing per query, and works on a plane. The trade-off: a 7B local model is not as capable than GPT-4 on hard reasoning tasks. If you need frontier capability, use a frontier tool. If you need a fast, private, always-available AI for daily work, Pulse is the better fit.
Pulse for Android v0.6.x uses Android’s on-device speech recognizer (free, accurate, no network round-trip). Desktop Pulse uses Whisper.cpp running locally on the device — the same model that powers the OpenAI Whisper API, but offline. There is no cloud STT in either surface.
Pulse for Android is a 1.29 MB APK, no Play Store, side-load from the link below. Pulse for Windows is a ~6.4 MB Tauri 2 NSIS/MSI installer, also from this site. Both are open source, both work offline, both are the same Apache 2.0 source tree.