What are the best open alternatives to Meta Muse?
nanoMuse is the closest open alternative to Meta Muse: a personal agent in Muse's style for your phone, computer and browser, GPL-3.0, on any OpenAI-compatible model. Rakazo adds iMessage and WhatsApp bridges and names Muse and Instinct as what it replaces, while OpenClaw and Hermes Agent are the mature personal agents with apps and chat channels everywhere. Meta Muse and Instinct can't use your own model or API key. Muse is still far easier: no setup, a managed computer, WhatsApp built in and Meta's shopping partners.
The open alternatives that take your own API
Each one documents a custom OpenAI-compatible endpoint in its own docs. Ordered by how close they come to Muse: a personal agent on your phone that does things for you and keeps working while the app is closed.
1. nanoMuse
An open-source personal agent "in the style of Meta's Muse" from Zhejiang University: one agent with a name and a look of its own that keeps working while the app is closed, remembers you and asks before anything you cannot undo, on Android, iPhone, iPad, desktop and the web. Add any OpenAI-compatible endpoint under Settings → Providers or the web console's Connections. GPL-3.0, non-profit, v1.0.5.
Wrong pick when you want an App Store install on iPhone. iOS is TestFlight only, and signing in uses a community relay at nanomuse.cn unless you host your own or use your own key.
2. Rakazo
Names Meta Muse, OpenAI Dots, Grok Bot and Instinct as what it replaces. Persistent bots with their own computers, apps for the web, desktop, iOS and Android, and bridges into iMessage, WhatsApp, Slack and Telegram, so you can text your bot like Instinct. Apache-2.0, beta.
Wrong pick when you don't want to run a server. You host the Docker Compose stack yourself, and the operator must allow public model endpoints in .env first.
3. OpenClaw
The open-source personal agent that inspired Muse: Meta's Nat Friedman said Muse was built from scratch but "heavily inspired as a product by openclaw". It has apps for macOS, iOS, Android, Windows and Linux and more than 20 chat channels, including WhatsApp and iMessage. A custom provider takes any OpenAI- or Anthropic-compatible base URL. MIT, about 392,000 GitHub stars.
Wrong pick when you want something you never have to maintain. You run its gateway on a computer or server that stays on, and the setup is aimed at technical users.
4. Hermes Agent
Nous Research's self-improving agent with long-term memory, in the terminal, Hermes Desktop, Telegram, WhatsApp, Signal, Slack and Discord. hermes model → Custom endpoint takes any OpenAI-compatible base URL. MIT.
Wrong pick when you want a phone-first app. Hermes lives in chat apps and on a computer you keep running.
5. OpenMuse (CopilotKit)
A personal agent with a browser, terminal, files and work that keeps going, with iOS, Android and web apps. MODEL, OPENAI_BASE_URL and OPENAI_API_KEY in .env set the model; set OPENAI_CHAT_COMPLETIONS=true for an endpoint without the Responses API. MIT.
Wrong pick when you don't want a third-party account. Every mode needs a CopilotKit Intelligence key and it is built for a single owner.
6. nanobot
HKU's lightweight self-hosted personal agent in Python, with a WebUI, memory, scheduled tasks and channels from Telegram to email. A named provider with apiBase in ~/.nanobot/config.json takes any OpenAI-compatible endpoint. MIT, about 49,000 GitHub stars.
Wrong pick when you want a polished phone app. nanobot is a framework you run and reach through its WebUI or chat apps.
How we chose
We searched GitHub and the web for projects that present themselves as Meta Muse or Instinct alternatives or that people compare with them, then kept only those whose own docs show a custom OpenAI-compatible endpoint and that carry an open-source license. Stars are GitHub counts on October 15, 2026. We left out open-dots by Anil-matcha (no license file and no custom endpoint) and Codync (it wraps other agents and has no endpoint of its own).
Sources: nanoMuse: bring your own key; Rakazo self-hosting guide; OpenClaw custom providers; Friedman on Muse and OpenClaw (Forklog); Hermes Agent providers; OpenMuse .env example; nanobot providers
Compared side by side
The custom endpoint column quotes each project's own docs. Meta Muse and Instinct are listed for scale.
| Project | License, GitHub stars | Runs on | Custom endpoint, from its docs | Also needs |
|---|---|---|---|---|
| nanoMuse | GPL-3.0 · 1.3k stars | Android, iPhone and iPad (TestFlight), Windows, macOS, Linux, web. Feishu, DingTalk, WeCom, Telegram | Phone: Settings → Providers → Add Provider. Web console: Connections → Chat model → Add an endpoint. Own runtime: [llm] with provider = "openai", base_url, model, api_key in config.toml | Nothing with your own key. Signing in uses the nanomuse.cn relay, which you can self-host |
| Rakazo | Apache-2.0 · 4.1k stars | Web, desktop, iOS, Android. iMessage, Slack, WhatsApp, Telegram and Feishu bridges | RAKAZO_OPENAI_COMPAT_ALLOW_PUBLIC=1 in .env, then Settings → Models → OpenAI-compatible | A Docker Compose server |
| OpenClaw | MIT · 392k stars | macOS, iOS, Android, Windows, Linux. WhatsApp, iMessage, Telegram, Slack, Discord and 20+ more | Onboarding wizard → More… → Custom Provider, or models.providers.<id> with baseUrl, apiKey, api: "openai-completions" in ~/.openclaw/openclaw.json | Node 24 |
| Hermes Agent | MIT · 254k stars | Terminal and Hermes Desktop. Telegram, Discord, Slack, WhatsApp, Signal | hermes model → Custom endpoint, or provider: custom with base_url in ~/.hermes/config.yaml | Nothing else |
| OpenMuse (CopilotKit) | MIT · 5.1k stars | iOS, Android, web | MODEL=openai/<id>, OPENAI_BASE_URL, OPENAI_API_KEY and OPENAI_CHAT_COMPLETIONS=true in .env (it defaults to the Responses API) | A CopilotKit Intelligence key in every mode |
| nanobot | MIT · 49k stars | WebUI and terminal. Telegram, Discord, Slack, WeChat, Feishu, Teams, email | WebUI Settings → Models, or a provider with apiBase and apiKey in ~/.nanobot/config.json | Python 3.11 |
| Meta Muse (closed) | Proprietary | iOS, Android, Mac, iPad, muse.ai, WhatsApp | None: Muse Spark only, no model or key setting | Free; Power $20 and Maximum $100 a month (reported) |
| Instinct (closed) | Proprietary | iMessage, WhatsApp, phone calls, Mac app, web | None documented | Invite only; free for now |
Stars: GitHub, October 15, 2026. Sources: each project's README and docs (linked in the first column and under the list above); Meta's Muse announcement; Muse's paid tiers as reported by TechCrunch; instinct.com. Any of these projects: base URL https://api.stdcmpt.com/v1 (Anthropic Messages: https://api.stdcmpt.com), your Standard Compute key, and the model standardcompute.
What Meta Muse is, and why it can't use your own model
Meta launched Muse on September 8, 2026 as "a secure, private personal AI agent that proactively helps with people's goals and suggests ideas", on iOS, Android, muse.ai and WhatsApp (Meta). A Mac app followed on September 18 and iPad on October 7. Each person gets a managed virtual machine with its own browser, and a separate agent checks what Muse sends to the internet. It launched in the US only.
Meta says Muse is free for most of what people need, with paid plans for more; TechCrunch reports Power at $20 and Maximum at $100 a month (TechCrunch).
Muse runs on Meta's own Muse Spark model, and its settings have no model picker, API key or base URL. Meta does sell Muse Spark as a model through its Model API (a US-only preview), but that gives you the model, not the Muse agent.
Instinct: the same limit
Instinct, from Spear Street Technology, is "a personal assistant that understands what you're working on" that you text or call, and it can use a phone and a computer the way people do (instinct.com). You reach it through iMessage, WhatsApp, calls, a Mac app and the web, and since September it has its own email address (TechCrunch). It is invite-only and free for now.
Instinct does not say which models it uses, and it has no model or API key setting. Its privacy policy says it uses third-party AI model providers and that interactions are used for training unless you opt out. Rakazo, OpenClaw and Hermes Agent are the open options you can text through iMessage, WhatsApp or Signal.
Where Muse and Instinct are better
- No setup. You install the app or send a text. Every open option here needs a server, Docker or a computer that stays on, and someone to keep it updated.
- A managed, isolated computer per person with a separate agent that reviews what goes out. With the open projects, sandboxing is whatever you configure.
- Muse works inside WhatsApp, can use what you saved on Instagram, and pays through Stripe Link with retail partners such as Shopify, Walmart and Instacart. Instinct can call businesses for you and has its own email address. We found nothing equivalent documented in the open projects.
- Polished apps on every platform. nanoMuse's iPhone app is still on TestFlight, and OpenMuse's apps are a template.
- On privacy the open options can do better: Muse and Instinct use interactions for training unless you opt out, while a self-hosted agent sends your data only to the model API you choose.
Run them through one API
A personal agent that checks your inbox, plans and runs routines makes many model calls a day. Point it at one OpenAI-compatible endpoint and you get one bill and one key to rotate.
- Cost first: send the model
standardcomputeand the router sends each step to the cheapest model that can do it. Routine steps such as reading pages, filling forms, summarizing and scheduled check-ins can go to low-cost models, and Claude Fable 5.1 or GPT-6 Astra handle the steps that need them. An exact model ID, such asanthropic/claude-sonnet-5.5, always stays on that model. - For every model we vet providers for price, uptime and speed, with automatic fallback when one fails, so a bot that runs overnight does not stop because one host is down.
- Opt-in Compliance rules restrict a key, team or company to provider endpoints in Europe or the US. They cover where the model runs. The bot's computer runs wherever you host the project.
- We don't train on your prompts or store them.
- Plans start at $19 a month for a fixed compute budget. At the end of the budget requests stop with HTTP 402 until renewal, so a looping bot cannot run up a bill.
Set up nanoMuse on Standard Compute
- In the web console open Connections → Chat model and, under Local or your own endpoint, click Add an endpoint. On the phone: Settings → Providers → Add Provider.
- Enter
https://api.stdcmpt.com/v1as the Base URL and your Standard Compute key as the API key. Leave Models (optional) empty, since the endpoint lists its models, and click Add endpoint. - Pick
standardcomputeas the chat model and save. The screen-operating model needs one that sees images, such asanthropic/claude-sonnet-5.5. - Rakazo, OpenClaw, Hermes Agent, OpenMuse and nanobot: the same base URL, key and model in the fields listed in the table above.
The catch
An open personal agent is a project, not an app. You host it, keep it updated and decide what it may touch, and an agent with your email, browser and accounts can do real damage if a prompt injection lands. Several of these projects are weeks old. Start with read-only access, and keep Muse or Instinct if zero setup matters more than choosing your model.
Where Standard Compute fits
One OpenAI- and Anthropic-compatible key for your personal agent, with a router that sends each step to the cheapest model that can do it, providers vetted for price, uptime and speed, opt-in Europe or US rules and no training on prompts. Plans start at $19 a month.
FAQ
Is there an open-source version of Meta Muse?
Not from Meta: the Muse agent and the Muse Spark model are closed. The closest community projects are nanoMuse (GPL-3.0, a Muse-style personal agent for phone, desktop and web), Rakazo (Apache-2.0) and CopilotKit's OpenMuse (MIT, needs a CopilotKit key). All three run on a model you choose.
Can Meta Muse use my own API key or another model?
No. Muse runs on Meta's Muse Spark and has no model picker, API key or base URL setting. Meta sells Muse Spark separately through its Model API, but that is the model, not the agent.
Is there an open alternative to Instinct?
No project copies Instinct directly, but Rakazo names it as what it replaces, and Rakazo, OpenClaw and Hermes Agent can all be texted through iMessage, WhatsApp or Signal while running on your own model. None of them documents Instinct's calls to businesses.
Which Meta Muse alternative works on iPhone?
OpenClaw and Rakazo have iOS apps. nanoMuse runs on iPhone and iPad through TestFlight, and CopilotKit's OpenMuse is a React Native template you build yourself.
Which API should I use for a self-hosted personal agent?
One that keeps a busy agent cheap and the bill fixed. With Standard Compute's model standardcompute, the router sends each step to the cheapest model that can do it, and requests stop with HTTP 402 at the plan's compute budget instead of billing overage. Plans start at $19 a month.
Is a self-hosted alternative more private than Muse?
It can be. Muse and Instinct use interactions for training unless you opt out. A self-hosted agent sends your data to the model API you configure, so pick one that doesn't train on prompts: Standard Compute doesn't train on them or store them, and an opt-in Europe or US Compliance rule limits which provider endpoints may serve the key.