sabato 21 novembre 2026

🐋 go-bot Review: A Self-Hosted Personal Agent Written in Go

go-bot whale logo

go-bot · Go implementation of DeepSeek Harness · self-hosted personal agent

🐋
go-bot: A Self-Hosted Personal Agent Written in Go

go-bot is an open-source Go implementation of DeepSeek Harness, packaged as a self-hosted personal agent. The project positions itself as a self-hosted alternative to Meta Muse, OpenAI dots, and Grok Bot — and after digging into the repo and the docs, it's clear this is not a toy: it's a serious multi-component system with a real architecture behind it.

✨ Why it caught my eye

The idea of a personal agent you run on your own infrastructure is the whole point: your data, your models, your rules. With 211 stars on GitHub and a solid MIT license, it's grown into a project with a defined architecture, a plugin marketplace, a mobile app, and even a demo using Stanford HAI 2026 AI Index data.

⭐ Key features

  • 🌐 Web research & browser automation — search the web via a configured provider, then use Chromium to navigate pages, fill forms, and capture screenshots directly.
  • 🎨 Image generation & editing — generate images from text or use uploaded images as references; results appear straight in the chat.
  • 📄 Office documents — read, create and edit Word, PowerPoint and Excel files via OfficeCLI, delivered as downloadable attachments.
  • 🧠 Long-term memory — save user preferences and agent notes per account, and search past conversations when a task needs earlier context.
  • 📱 Mobile client — an Android APK plus an iOS/web flow that connect to the same Controller over Tailscale.
  • 🔌 HTTP + SSE API — the built-in web UI is a test client; the real power is in the Controller's API for building your own frontends.

⚙️ How it works (in 30 seconds)

  • 1️⃣ Spin it up — install Docker + Compose, run ./scripts/start.sh --init-env then ./scripts/start.sh from the repo root.
  • 2️⃣ Create an account — open http://127.0.0.1:8080, create an account by name; opening it starts its own Host container.
  • 3️⃣ Configure a model — pick OpenAI Responses or Anthropic Messages in Model settings, add endpoint URL, API key and a model ID.
  • 4️⃣ Enable plugins — install and enable agent-browser and OfficeCLI from the Plugins market to unlock browser and Office tasks. No API key needed for those two.

🧪 My honest take

What makes go-bot stand out is its architecture: a Controller that serves the web UI, manages Hosts and forwards HTTP/SSE, with Redis storing accounts and state. Each account gets its own Host — a Go agent running Chromium, Python, Node.js and the document environment, managing tools, session journals, memory, schedules and attachments. That per-account isolation is a design choice you don't see in simpler agent projects.

The deployment options are equally interesting: local Docker for personal use, and Alibaba Cloud Elastic Container Instance (ECI) for a team-scale "personal agent product" with one dedicated Host per account, optional NAS persistence, and a built-in image cache workflow. That's a rare combination — self-hosted simplicity with a real enterprise path.

🏆 The verdict

go-bot is one of the more serious self-hosted agent projects I've looked at recently. It's not just a chat wrapper: it bundles memory, scheduling, browser automation, document creation and image editing into one coherent system, and it stays yours. If you want your own personal agent without renting it from a big platform, this is worth a spin. The main trade-off: it's a Docker-first setup and the ECI path assumes Alibaba Cloud, so it's most approachable if you're comfortable running containers locally.

🚀 Ready to run your own agent?

👉 The project is on GitHub:

https://github.com/xyc999-ai/go-bot

Readme in 简体中文

❓ FAQ

🤔 Do I need a cloud provider to use it?

No. The default path is local Docker: Controller + Redis via Compose, and each account gets its own Host container. Alibaba Cloud ECI is an optional deployment for teams that want one dedicated Host per account with NAS persistence.

🤔 Which models can I connect?

You configure the model provider yourself in Model settings — OpenAI Responses or Anthropic Messages protocol, with your own endpoint URL, API key and model IDs. Web search providers include DeepSeek, Zhipu, Doubao, Tavily or Brave Search.

🤔 Does it work on my phone?

Yes. An Android APK is available in the GitHub Releases assets, and iOS users can reach the web UI via Tailscale. Both connect to the same Controller over a private tailnet, which works without public HTTPS certificates.

🤔 What about office documents?

Install and enable OfficeCLI from the plugin market and the agent can read, create and revise Word, PowerPoint and Excel files, delivering them as downloadable attachments. It ships with 11 original skills.

🤔 What license is it under?

go-bot is released under the MIT license, with a security policy and third-party notices included in the repo.

I regali che desideri davvero

Quando ti chiedono «Cosa vuoi per il compleanno?», ti vengono in mente solo le idee che avevi dimenticato di salvare? Con RegaloNoProblem puoi raccogliere i tuoi desideri man mano che li trovi e condividere un unico link con amici e parenti.

✨ Cosa funziona bene

La homepage spiega in tre passaggi come funziona la lista. La funzione più interessante è la prenotazione dei regali. Se qualcuno ha già scelto un articolo, gli altri lo vedono e possono orientarsi su un altro regalo. Un dettaglio pratico quando più persone fanno acquisti per la stessa occasione.

  • 🔗 Aggiungi un desiderio. Incolla il link al prodotto. Se il negozio mette a disposizione una foto, questa può comparire nella scheda.
  • 💌 Condividi la lista. Basta mandare il link. Chi lo riceve può aprirlo senza un account.
  • ✅ Evita i doppioni. Gli invitati possono prenotare un regalo direttamente dalla lista condivisa.

🎉 Hai già un paio di idee da salvare? Crea la tua lista su RegaloNoProblem.

❓ Domande frequenti

👥 Serve un account per aprire una lista?

Chiunque abbia il link condiviso può aprire la lista senza registrarsi.

🖼️ Ogni regalo avrà una foto?

Dipende dal negozio. La foto può comparire nella scheda se il negozio la rende disponibile.

🎀 Cosa succede quando qualcuno prenota un regalo?

Gli altri invitati vedono che l’articolo è già stato scelto e possono sceglierne un altro.

Give Friends a Clue They’ll Actually Use

“What would you like?” is an easy question to ask and a surprisingly hard one to answer on the spot. RegaloNoProblem lets you collect gift ideas as you find them, then share a single wishlist when someone asks.

✨ What Works Well

The homepage explains the whole process in three steps. The standout feature is gift reservations. Once a guest reserves an item, everyone else viewing the list can see it’s already spoken for. That’s especially handy when several people are shopping for the same occasion.

  • 🔗 Add an idea. Paste a product link. If the shop provides a photo, it can appear on the gift card.
  • 💌 Share your wishlist. Friends and family can open it from the link without an account.
  • ✅ Keep gifts from doubling up. Guests can reserve items directly from the shared list.

🎉 Got a birthday coming up or a few gift ideas worth saving? Try RegaloNoProblem and create your wishlist.

❓ Frequently Asked Questions

👥 Do guests need an account?

Anyone with the shared link can open the wishlist without an account.

🖼️ Will every gift have a photo?

Photos depend on whether the shop provides one for the product you link to.

🎀 What happens when someone reserves a gift?

Other guests can see that the item has already been chosen, so they can pick something else.

domenica 15 novembre 2026

🚀💸 Actual Budget: The Privacy-First Budgeting Revolution 💸🚀

👋 Hey Techies! Are you still paying monthly subscriptions just to track your expenses? Do you hate the idea of your bank data sitting in a corporate cloud? If you nodded "yes" to both, you need to meet Actual Budget.

This isn't just another spreadsheet; it's a self-hosted, privacy-focused powerhouse that makes managing money feel like a game, not a chore.

🌟 Why It’s a Game Changer

Here is why tech-savvy money managers are switching to Actual:

  • 💰 True Zero-Based Budgeting: Every dollar gets a job. It uses the classic "envelope method" but with the power of a modern app.
  • 🔒 Local-First & Private: Your data lives on your machine, not in a corporate database. You own your financial truth.
  • 💸 Open Source & Free: No hidden fees, no "Pro" versions locked behind paywalls. Just pure budgeting power.
  • 🔄 Sync On Your Terms: Need to sync between devices? You can set up sync yourself or use their optional paid sync service. No forced subscriptions.

🧐 The Verdict

Actual Budget bridges the gap between the ease of Mint/YNAB and the control of Excel. It’s robust, customizable, and respects your privacy.

👇 Ready to take control of your finances?

Check out the project on GitHub and start budgeting your way!

🚀 Visit Actual Budget on GitHub

❓ Frequently Asked Questions

❓ Is Actual Budget really free?
Yes! The software is open-source and free to use. Syncing across devices is optional and has a small fee if you use their service, but you can also self-host for free.

❓ Is it secure?
Since it is "Local-First," your data stays on your device by default. This means even if the developers' servers go down, you still have full access to your data.

❓ Does it have a mobile app?
Yes! It works beautifully as a Progressive Web App (PWA) on mobile, or you can use community-built apps like Actual Budget Mobile.

martedì 3 novembre 2026

LibreCorsa — Simulatore di motorsport open-source nel browser

LibreCorsa è un simulatore di guida open-source che gira direttamente nel browser, senza installazioni. Progetto realizzato con Next.js, React e Three.js (licenza MIT), punta dritto alla fascia alta dei simulatori web con fisica a 500 Hz.

Cosa c'è sotto il cofano

  • Fisica: integrazione a passo fisso 500 Hz (con 4 sub-step a 2 kHz), corpo rigido 3D con ogni ruota che segue la reale superficie di pista
  • Pneumatici: modello termico a 9 nodi (Pacejka-style slip curve, combined slip, carcass relaxation), usura dei battuti. Soft, medium, hard slicks + intermedi e full wets con modello aquaplaning
  • Telaio: doppio triangolo di leve, barre anti-rollio, ammortizzatori, limit-slip differenziale (preload/power/coast), freni termici

Vetture

  • Open-wheeler 2026: V6 turbo 1.6L + MGU-K da 350 kW, 4 MJ batteria, Z/X-mode aerodinamico
  • Mercedes-AMG GT3: V8 NA 6.2L, cambio sequenziale a 6 marce, ABS multi-stage, controllo trazione, modello braking-point dedicato

Piste e funzionalità

4 circuiti inclusi (tra cui Shanghai 2018 e Suzuka 2001), gare con IA, fantasmi e replay, multiplayer peer-to-peer. Supporta wheel con force feedback via WebHID.

Verdetto

La fisica è impressionante per un simulatore web — non gira nel cloud ma localmente nel browser. Nessun server necessario, zero installazioni. Per chi vuole testare setup e vettura da desktop senza scaricare gigabyte di software, è una soluzione pulita. Da seguire: il progetto è giovane (16 star su GitHub) ma la qualità del codice e delle meccaniche fa ben pensare per gli sviluppi futuri.

Provato su: khazp.github.io/librecorsa

mercoledì 28 ottobre 2026

One Keyboard, One Mouse — Control Every Machine Seamlessly

🖥️

One Keyboard, One Mouse — Control Every Machine Seamlessly


Tired of switching between your Mac, Windows PC, and Linux boxes? Beamer lets you use a single keyboard and mouse across all your machines on the same network. Push your cursor off the edge of one screen and it flows right onto the next — just like a multi-monitor setup, but with real computers.


✨ What It Does

  • 🖱️ Cross the edge — push your pointer through the screen border and it arrives on the other machine at the same position
  • ⌨️ Keyboard follows you — type natively on any connected machine with smart key remapping (Cmd↔Ctrl, Option↔Alt)
  • 📋 Clipboard travels — copy text and images on one machine and paste them on another instantly
  • 🎯 Mac gestures on Windows — swipe your trackpad for Task View, desktop switching, and more
  • 🔒 Your network only — no cloud servers, encrypted P2P connection, paired with a simple 6-digit code


⚡ Quick Setup

  • 🍎 macOS: Download the .dmg, drag to Applications, grant permissions and relaunch
  • 🪟 Windows: Run the installer — approve SmartScreen warning with "More info" → "Run anyway"
  • 🐧 Linux (Flatpak):
flatpak install --user https://flatpak.kalkmancode.co.uk/beamer.flatpakref


💬 FAQ

Is Beamer free? Yes — completely free, open-source under GPL-3.0, no account required.

Can I connect more than two machines? Absolutely! Pair each machine with every other one for a chain of any size.

Does it work between two Macs or two PCs? Yes — Beamer works in any direction and between identical platforms too.


👉 Give it a try today — your workflow will thank you!

🔗 Check out Beamer on GitHub

lunedì 26 ottobre 2026

🔊 Pocket TTS: Speech That Fits in Your Pocket

Tiny TTS for Any CPU

Forget about GPUs and heavy web APIs. Pocket TTS turns text into speech with just a pip install and a function call.

Why Pocket TTS?

  • 🚀 Runs on plain CPUs — no GPU required
  • 🪶 Ultra-small models, down to 1.9M parameters
  • ⚡ Low latency, ~200ms for the first audio chunk
  • 🧠 Supports voice cloning and multi-language (english, french, german, italian, etc.)

Try It with the CLI

Use the generate command directly from the command line:

...pocket-tts generate --text "Hello world." --voice marius...

Adjust the voice with --voice and text with --text. Choose a pretrained language model with --language when running generate, export-voice, or serve (default: english).

Quick Tip: Streaming & Server

Generate audio instantly over HTTP with the serve command:

...pocket-tts serve...

Navigate to http://localhost:8000 for a faster web interface where the model stays in memory between requests.

FAQ

Q: Does it work on a laptop CPU?

A: Absolutely — it's designed for CPU-only, even 1.9M parameter models run fast.

Q: Can I use custom voices?

A: Yes, via voice cloning or by providing your own audio files directly.

Q: Is it easy to install?

A: Just run pip install pocket-tts or use uvx pocket-tts — no heavy setup needed.

Get Started Today

🎙️

🔗 https://github.com/kyutai-labs/pocket-tts

giovedì 22 ottobre 2026

🚀 workout.cool: The Modern Fitness Coaching Platform

🚀 workout.cool: The Modern Fitness Coaching Platform

🎯 A Comprehensive Fitness Platform

A modern fitness coaching platform that lets you create workout plans, track progress, and access a vast exercise database with detailed instructions and video demonstrations.

💡 The Story Behind workout.cool

This project was born to revive what we lost. As the primary contributor to the original workout.lol, I witnessed its abandonment after video licensing costs became prohibitively expensive. Rather than let this valuable work disappear, I created a fresh, modern implementation — an evolution, not just a revival.

👥 Built For You

Whether you're a fitness enthusiast or a developer, this platform is built for the community. Track your progress, manage your routines, and explore a comprehensive exercise database with rich video demonstrations.

📱 Quick Start

🐳 With Docker
  1. Copy environment variables
  2. Start everything for development
  3. cp .env.example .env
    make dev
  4. Open your browser
💻 Without Docker
  1. Install dependencies
  2. Copy environment variables
  3. Start the development server
  4. pnpm install
    cp .env.example .env
    pnpm dev

🔗 Join the Rescue Mission

Star this repo, join our Discord, report issues, and share feature requests — we're listening to every single one.

🚀 [https://github.com/Snouzy/workout-cool](https://github.com/Snouzy/workout-cool)

❓ FAQ

Q: How do I install workout.cool?

A: Clone the repository, run pnpm install, and copy .env.example to .env.

Q: Can I use Docker to run it?

A: Yes! Use the Docker build and run commands to deploy with a PostgreSQL database.

Q: What license applies to this project?

A: It's licensed under the MIT License — free to use and modify freely.

heartlib: Open-Source Music Foundation Models

🎵

heartlib: Open-Source Music Foundation Models

The most powerful open-source music generation model of 2026

🎶

A family of open-sourced music foundation models

  • 🎼 HeartMuLa — music language model generating music conditioned on lyrics and tags with multilingual support
  • 🎧 HeartCodec — 12.5 Hz music codec with high reconstruction fidelity
  • 📝 HeartTranscriptor — whisper-based model tuned for lyrics transcription
  • 🔗 HeartCLAP — audio–text alignment model for unified embedding space
🚀

Latest releases — 7B and 3B versions

  • 🔥 HeartMuLa-7B — comparable to Suno in musicality, fidelity and controllability
  • 🎬 MuLaCover — controllable cover-song and music-remix model
  • 🎧 HeartCodec-oss-encoder — audio tokenization and reconstruction
🖥️

Local deployment in one step

... git clone https://github.com/HeartMuLa/heartlib.git

... pip install -e .

⚖️

License & citation

... Apache 2.0 License

❓

FAQ

Q: How do I generate music?

A: ... python ./examples/run_music_generation.py --model_path=./ckpt --version="3B"

Q: What versions are available?

A: ... HeartMuLa-3B and HeartMuLa-7B (7B being the latest)

Q: How do I install it locally?

A: ... git clone ... / pip install -e .

🎵 Start generating music with heartlib today!

... https://github.com/HeartMuLa/heartlib

mercoledì 21 ottobre 2026

colibri: Tiny Engines, Immense Models

🐦

colibri: Tiny Engines, Immense Models

Run frontier MoE models on hardware you already own

(/)

A mixture-of-experts model of hundreds of billions of parameters

Pure C, zero dependencies, experts streamed from disk.

🧠

13 engines ready — from 35B to 2.8T parameters

  • 🐦 GLM-5.2 / 5.3 (744B MoE)
  • 🤖 DeepSeek V4 Flash & V4.1 (284B / 552B)
  • 🔥 Qwen3.8-Flash-Next (125B)
  • 🎨 Qwen-Image-2.1 (text to picture)
  • 🧩 Decision models: Laya, GLiNER2.5, Clef
🖥️

No GPU required — CPU runs everything, GPU accelerates

  • 🖥️ CPU: every engine runs with nothing else installed
  • 🎮 Vulkan: any GPU with 1.2 driver
  • 🚀 CUDA: NVIDIA cards with toolkit support
  • 📦 Multi-drive: read experts from NVMe, two drives +37.5%
⚡

One step to get started

... ./start-here.sh — asks for your machine specs, recommends a model, builds or downloads the engine, and opens the dashboard.

... ./start-here.sh --yes — no questions, go straight to the recommendation.

🧩

System One: make models answer choices, not paragraphs

... POST /v1/systemone — gives the model a document and only allowed options, returning probabilities and confidence.

❓

FAQ

Q: Can I run large models on a laptop?

A: Yes — on a 25 GB laptop, every expert streams from disk and runs correctly, though slower.

Q: Do I need a GPU?

A: No — every engine runs on CPU; a GPU is optional but faster for the hot experts.

Q: How do I connect colibri to other apps?

A: Yes — use the OpenAI-compatible API (base URL http://127.0.0.1:8000/v1) or the Anthropic-compatible endpoint; existing clients work unmodified.

🎉 Get started now and watch your own hardware run frontier models!

... https://github.com/JustVugg/colibri

🐋 go-bot Review: A Self-Hosted Personal Agent Written in Go

go-bot · Go implementation of DeepSeek Harness · self-hosted personal agent 🐋 go-bot: A Self-Hosted Personal Agent Written in Go ...