Hope Agent: Cross-Device Desktop AI Agent
/hope-agent
Hope Agent is a desktop AI agent built in Rust that keeps working toward a Goal after you close the chat window, instead of answering one prompt at a time. It pairs a native macOS GUI with an optional headless server mode for a NAS or cloud host. Reach for it if you want an agent that runs multi-step Goals and Workflows in the background; skip it if you need solid Windows or Linux support, since the README lists both as experimental.
Introducing Hope Agent: A Cross-Device AI Assistant
Hope Agent is a local-first, desktop-first personal AI agent from developer shiwenwen, written in Rust with a Tauri desktop shell and a React 19 frontend, that can also run as a headless server. The README describes it holding long-term context, calling tools with your approval, and continuing to push a Goal forward after you step away, rather than only responding to single chat messages like a typical assistant.
Comprehensive AI Agent Capabilities
- ✓A Goal/Workflow/Loop model for long-running tasks: a Goal sets the outcome, a Workflow dynamically arranges steps (stages, conditions, parallel branches, multi-agent, tools, diff review), and a Loop decides when to resume — work can run in the background and be paused, resumed, or checked on later, per the README.
- ✓Layered long-term memory across global, project, and agent scopes, with a compact 'Core' kept in context and full detail pulled back in via full-text and vector search on demand, plus an idle-time review that extracts communication style and habits from history, per the README.
- ✓A Knowledge Space where you and the agent read and write real Markdown notes together, with full-text and vector search, backlinks, a graph view, and an optional bind to an existing Obsidian vault, per the README.
- ✓A Design Space that turns a sentence or reference image into 10+ deliverable types (web pages, mobile prototypes, presentations, dashboards, posters, documents, images, motion, audio) with visual tweaking, version history, and export to HTML/PNG/PDF/PPTX/MP4/ZIP, per the README.
- ✓Computer and browser control under approval — the agent can operate windows, menus, keyboard, and mouse on macOS, and drive a browser with a live mirror of what it's doing, with side-effecting actions gated behind approval, per the README.
- ✓A provider ecosystem the README puts at 40+ templates and 300+ preset models, spanning Anthropic, OpenAI, Gemini and other international providers, Chinese providers like DeepSeek and Moonshot (Kimi), and local models through a one-click Ollama install.
- ✓Cross-device handoff: Desktop, Server, Web, and ACP modes share the same ha-core logic, so a session's memory and task state can continue across desktop, browser, and connected IM channels like Telegram, Discord, Slack, or Feishu, per the README.
Installation and Deployment Options
Hope Agent ships prebuilt installers for macOS, Windows, and Linux from GitHub Releases, mirrored at repo.hopeagent.ai. On macOS the README recommends Homebrew (`brew tap shiwenwen/hope-agent` then `brew install --cask hope-agent`) or a manual DMG, with a documented `xattr`/`codesign` workaround if macOS flags the app as damaged or from an unverified developer. Windows installs via Scoop (`scoop bucket add hope-agent https://github.com/shiwenwen/scoop-hope-agent` then `scoop install hope-agent`) or a manual .exe installer, x64 only. Linux is covered by an AUR package, an apt repo for Debian/Ubuntu, a dnf/yum repo for Fedora/RHEL/CentOS, a zypper repo for openSUSE, or manual AppImage/deb/rpm downloads, with amd64 and arm64 builds. To self-host, run `docker run` against the `ghcr.io/shiwenwen/hope-agent:latest` image with a mounted data volume and port 8420 open; the container bundles the Web GUI. First launch walks you through picking a provider template and entering an API key or Codex OAuth login.
Strengths
- ✓Local-first by default: the README states config, sessions, memory, attachments, skills, and logs are saved under `~/.hope-agent/`, with API keys going straight to the model provider instead of through a relay.
- ✓A genuinely broad provider list for one tool — 40+ templates and 300+ preset models spanning Anthropic, OpenAI, Gemini, DeepSeek, Moonshot (Kimi), Qwen, and more, per the README, so you're not locked into one vendor.
- ✓The Goal/Workflow/Loop split gives long tasks a real state machine — pause, resume, and background execution are described as first-class, not an afterthought bolted onto a chat loop.
- ✓Approval-gated computer and browser control, with an optional Docker sandbox for high-risk commands and file writes, per the README.
- ✓Runs the same way on a NAS or VPS as on a laptop — `hope-agent server` embeds the full Web GUI in the binary, so you're not stuck with a desktop-only build.
Considerations for Hope Agent Use
- △The README marks Linux and Windows as experimental — only macOS is described as having the complete desktop experience today, and Windows specifically says it hasn't finished thorough testing yet.
- △Manual macOS installs can trigger a Gatekeeper warning that the app is damaged or from an unverified developer, requiring a documented `xattr`/`codesign` workaround rather than a notarized, double-click install.
- △The Windows build is x64 only and needs the Microsoft Visual C++ 2015-2022 redistributable installed separately if a missing `MSVCP140_1.dll` or `VCRUNTIME140.dll` error shows up, per the README.
- △One-click local model installation depends on Ollama's runtime underneath it — the README credits Ollama for the actual local inference, with Hope Agent acting as a GUI layer on top.
- △The README itself says a roadmap, an official docs site, and more community channels are still in preparation, so some of that structure isn't public yet.
- △No numbered release version appears in the provided facts, so there's no way to confirm from the README alone how mature the current build is.
Exploring Similar AI Agent Tools
Hope Agent Frequently Asked Questions
Hope Agent supports macOS with what the README calls the complete desktop experience, plus Windows and Linux, which the README currently labels experimental. The Windows build is x64 only; Linux ships amd64 and arm64 builds covering regular PCs, Raspberry Pi 4/5, and Apple Silicon running Asahi Linux.
Yes — the `hope-agent server` subcommand runs Hope Agent headless on a NAS, home server, or cloud host with the Web GUI embedded in the binary, and it can register as a launchd or systemd service to start on boot, per the README.
Yes, Hope Agent installs Ollama with one click from its settings, no account, API key, or terminal needed, and picks a model sized to your hardware; the README credits Ollama's runtime for the actual local inference, with the same flow used for local embeddings.
Hope Agent is released under the MIT license, a permissive open-source license that also allows commercial use and modification.
The README describes Hope Agent as local-first with no third-party relay: config, sessions, memory, attachments, skills, and logs are saved by default under `~/.hope-agent/`, and API keys connect directly to your chosen model provider. Server mode adds Bearer token authentication and an SSRF-protection policy, per the README.
Yes, the README lists 10+ IM channels including Telegram, Discord, Slack, Feishu, Google Chat, LINE, QQ Bot, Signal, iMessage, IRC, WeChat, and WhatsApp, with images, voice, and files entering the conversation directly and sessions handing off between desktop, browser, and IM.
The problem it solves
Most AI assistants live inside a chat window: you type a prompt, get an answer, and the work stops the moment you close the tab or run out of context. Hope Agent's README frames its own reason for existing around that gap — pushing the agent out of the chat box into something that holds long-term context, keeps working on a Goal after you leave, and picks the session back up wherever you next open it, on desktop, in a browser, or in an IM app you already use.
Best use cases
- •Running a multi-step Goal overnight (research plus a design deliverable, say) on the desktop app and checking progress from a phone over Telegram or Slack the next morning, since the README describes shared session and memory state across desktop and IM.
- •Standing up a headless Hope Agent on a home NAS or VPS via `hope-agent server` or Docker so it stays online without a GUI, then reaching it from any browser on the network.
- •Building a second brain in Markdown where you and the agent co-edit notes, with the agent recalling prior context instead of starting from zero each session.
- •Going from a one-line brief to a working prototype (web page, deck, dashboard) in the Design Space, then handing the chosen output to the main conversation to implement in a real code repo, per the README.
- •Using ACP mode (`hope-agent acp`) to plug Hope Agent into an ACP-compatible IDE as the agent backend instead of running the desktop app.
Who should try it — and who should skip
Try Hope Agent if you're on macOS and want an agent that keeps a Goal moving in the background — checking email, building a design deliverable, or working through a Workflow — while you're away from the keyboard, and you like the idea of picking it back up from your phone over IM. Skip it if you're on Windows or Linux and need a fully baked experience today, since the README calls those platforms experimental; skip it too if you're not comfortable handing an agent computer and browser control, even with the approval gate the README describes.
Related repositories
Still deciding about hope-agent?
One click hands the question to an AI along with this page — see what it says about hope-agent.