Project directory

Personal agents

Assistants that work across your messages, calls and apps.

Rakazo

Rakazo

Rakazo is a self-hostable workspace for assistants that run routines and work on their own remote desktops.

  • Deployment. Published images require Docker Engine and Compose. The default setup runs bot computers in local Docker containers. Source ↗
  • Triggers. Routines can run on schedules, incoming email or webhooks and expose a run history. Source ↗

Publisher update ·

Rakazo v0.13.0 ↗

Scheduled · awaiting first completed check

Grok Bot

SpaceXAI

Grok Bot is a hosted assistant that runs persistent personal or shared team workflows on a cloud computer.

  • Runtime. Bots share a persistent cloud computer with a browser, filesystem and terminal. Source ↗
  • Routines. A bot can save a completed process as a reusable skill. Routines can run skills on schedules or in response to events. Source ↗

Monitored

Agent Zero

Agent Zero

Agent Zero is a self-hosted assistant that works in a Linux desktop with browser and document tools.

  • Deployment. A desktop launcher for macOS, Windows and Linux starts Docker-based instances. Servers can use a shell installer. Source ↗
  • Model access. An instance needs a supported AI account, provider API key or local model server. Source ↗

Publisher update ·

Agent Zero v2.16 ↗

Scheduled · awaiting first completed check

OpenClaw

OpenClaw

OpenClaw is a self-hosted assistant gateway that connects model providers to messaging apps and device clients.

  • Gateway. The gateway owns sessions, tools, events and channel connections. Source ↗
  • Channels. Telegram, Slack, Discord and WhatsApp can connect to the same gateway. Source ↗

Publisher update ·

OpenClaw 2026.11.5 ↗

Scheduled · awaiting first completed check

OpenBot by CopilotKit

OpenBot by CopilotKit

OpenBot by CopilotKit brings built-in bots and externally hosted agents into a workspace with shared policies.

  • Deployment. The macOS and Windows desktop apps bundle a Docker-based backend. Web, mobile and chat access are in early access. Source ↗
  • Integration. AG-UI connects external agents and MCP connects tools. Source ↗

Publisher update ·

OpenBot v1.1.0 ↗

Scheduled · awaiting first completed check

Tasks in Gemini is a hosted mode in the Gemini app for delegating multi-step work across Google and third-party apps.

  • Access. Google announced a free rollout in the U.S., Japan, South Korea, Canada, Australia and India, with advanced features starting in the U.S. Source ↗
  • Entry points. The Gemini app has a Task toggle. U.S. users can also forward email or text Gemini with an eligible U.S. number and carrier. Source ↗

Publisher update ·

Gemini expands Tasks and starts a free rollout ↗

Scheduled · awaiting first completed check

Muse is Meta’s hosted personal agent for planning and carrying out tasks across connected apps and the web.

  • Cloud runtime. The agent and its browser run on a dedicated cloud VM, so work can continue after the app closes. Source ↗
  • Control. Users choose access per app. Muse asks before sensitive actions such as sending email or making a purchase and shows an audit trail. Source ↗

Publisher update ·

Muse launches in Australia ↗

Scheduled · awaiting first completed check

Poke

Poke

Poke is a hosted messaging assistant that connects personal apps, sets reminders and runs automations.

  • Channels. Poke lists Apple Messages, WhatsApp and Telegram as messaging surfaces. Source ↗
  • Integrations. Connected Gmail can search, draft and send email. Poke also lists Notion pages and databases as a supported integration. Source ↗

Publisher update ·

Devin Bots ↗

Scheduled · awaiting first completed check

LobsterAI

LobsterAI

LobsterAI is a desktop assistant that works with local files, runs recurring tasks and receives requests from chat apps.

  • Platform. Official downloads cover Apple Silicon and Intel Macs and x64 Windows. Computer use is documented for Windows x64 and macOS 13 or later. Source ↗
  • Model access. Built-in models require sign-in. Custom model settings accept the operator's own API key, and the app also supports local model servers such as Ollama and LM Studio. Source ↗

Publisher update ·

LobsterAI 2026.11.16 ↗

Scheduled · awaiting first completed check

nanobot

nanobot

nanobot is a self-hosted Python assistant that handles files and scheduled tasks across a web interface, terminal and chat apps.

  • Requirements. The runtime needs Python 3.11 or later and access to a supported model provider or local model server. Source ↗
  • Gateway. The gateway manages agent execution, tools, memory and session state for browser, terminal and chat clients. Source ↗

Publisher update ·

nanobot v0.4.1 ↗

Scheduled · awaiting first completed check

Lindy

Lindy

Lindy is a hosted team assistant that coordinates recurring work across Slack, email, calendars and connected business tools.

  • Delivery. Teams can use Lindy in Slack threads, over email or iMessage, and in the browser. Source ↗
  • Automation. Scheduled routines can draft replies, prepare meeting follow-ups and run recurring work with connected tools. Source ↗

Publisher update ·

Webhooks for Lindy Apps ↗

Scheduled · awaiting first completed check

Moltis

Moltis

Moltis is a self-hosted Rust agent server with a web interface, chat integrations and scheduled tasks.

  • Deployment. Moltis is MIT licensed and ships as a macOS or Linux binary and as a Docker image for amd64 and arm64. Source ↗
  • Model access. GitHub Copilot requires an active subscription. Ollama and LM Studio can connect to a local instance without an API key. Source ↗

Publisher update ·

Moltis 20261023.01 ↗

Scheduled · awaiting first completed check

NanoClaw

NanoClaw

NanoClaw is a self-hosted assistant that runs chat agents and scheduled jobs inside Linux containers.

  • Runtime. The host runs on macOS, Linux or Windows via WSL2. Onboarding uses Node.js, pnpm and Docker. Source ↗
  • Model access. Default setup uses the Claude Agent SDK and an Anthropic credential. Other provider integrations are installed as skills and can be configured per agent group. Source ↗

Publisher update ·

NanoClaw v2026.10.0 ↗

Scheduled · awaiting first completed check

IronClaw

IronClaw

IronClaw is a self-hosted assistant with browser and messaging interfaces, scheduled routines and isolated tools.

  • License. The source code is dual licensed under MIT or Apache 2.0. Source ↗
  • Model access. Guided setup asks for a model provider and its API key and stores the key in the local encrypted credential store. Source ↗

Publisher update ·

IronClaw 1.4.1 ↗

Scheduled · awaiting first completed check

OpenAI dots are hosted ChatGPT assistants that keep working across projects between conversations.

  • Access. Dots are rolling out across ChatGPT web, mobile and desktop on Pro, Business Premium and Enterprise in eligible markets. Enterprise access requires an admin to enable it. Source ↗
  • Runtime. A dot starts with ChatGPT memory, uses Codex and connected apps, and runs on its own cloud computer. Connecting a user's computer is optional. Source ↗

Publisher update ·

Introducing dots ↗

Scheduled · awaiting first completed check

NullClaw

NullClaw

NullClaw is a self-hosted personal assistant runtime that compiles to a standalone Zig binary.

  • License. The self-hosted runtime is MIT licensed. Source ↗
  • Model access. Model routes include API-key providers such as OpenRouter and Anthropic, plus custom OpenAI-compatible endpoints. Source ↗

Publisher update ·

NullClaw v2026.5.29 ↗

Scheduled · awaiting first completed check

Khoj

Khoj

Khoj is a self-hosted assistant for searching and chatting over personal documents and the web.

  • Hosted access. The public Khoj Cloud service shut down on April 15, 2026. The self-hosted software remains available. Source ↗
  • Model access. Self-hosted instances can run via Docker or Python and use a commercial model API key or a compatible local endpoint such as Ollama. Source ↗

Publisher update ·

Khoj 2.0.0-beta.28 ↗

Scheduled · awaiting first completed check

Hermes Agent

Nous Research

Hermes Agent is a persistent assistant that runs in a terminal or messaging gateway and can schedule tasks and reuse skills.

  • Model and tool access. Operators can supply their own provider credentials or use the Nous Portal subscription for hosted model and tool access. Source ↗
  • Platform. The self-hosted agent supports Linux, macOS, WSL2 and native Windows, with a separate Termux package for ARM64 Android. Source ↗

Monitored

Leon

Leon

Leon is a self-hosted personal assistant with modular skills, local or remote models and profile-specific memory.

  • Preview status. The develop branch offers an unfinished local web app preview. The project describes the first 2.0 desktop alpha as unreleased. Source ↗
  • Runtime. The source preview needs Node.js 24 or newer and pnpm on macOS, Linux or Windows. Source ↗

Scheduled · awaiting first completed check

Compare personal agents

  • Hermes Agent

    Where it runs
    Self-hosted, with a terminal or messaging interface.
    Model and account access
    Connect a supported model provider or a compatible local endpoint.
    What to try
    Turn a recurring task into a reusable skill.
    Latest verified publisher change
    No published change on this timeline yet.
    + Follow
  • Grok Bot

    Where it runs
    Hosted, with a persistent cloud computer for the bot.
    Model and account access
    Uses an eligible subscription. The publisher documents supported plans.
    What to try
    Assign a task and send a follow-up to refine it.
    Latest verified publisher change
    · Grok Bot 1.2.3 adds per-tool connector permissions and routine reconnect cards
    + Follow
  • Rakazo

    Where it runs
    Hosted or self-hosted, with sandbox options.
    Model and account access
    Connect a supported model provider. Account requirements depend on the setup.
    What to try
    Review a proposed message before the agent sends it.
    Latest verified publisher change
    No published change on this timeline yet.
    + Follow