A personal AI agent runs continuously, keeps memory across sessions, and acts inside your apps. You talk to it in a chat app, by email or by voice. It keeps working after you close the window.
The category started small. OpenClaw began in November 2025 as Clawdbot, a weekend project by Peter Steinberger. Twelve months later, Google, OpenAI, Meta and Perplexity all ship their own versions.
The products split into 2 camps. Self-hosted agents run on your machine or server and use any model you choose. Managed agents run on a vendor’s cloud VM and come bundled with a subscription.
How we ranked them
We scored each agent on 3 axes that matter once an agent holds your inbox and credentials.
- Memory: what persists, where it is stored, and whether you can read, edit or delete it.
- Security: default permissions, the isolation boundary, approval gates, and the public incident record.
- Cost: the subscription price, or token spend plus hosting for self-hosted options.
Facts come from vendor documentation, launch posts and independent security research. Prices are US list prices as of November 21, 2026. Ranking reflects overall fit for an individual user, not raw capability.
Comparison table
| Rank | Agent | Maker | Runs on | Memory | Security default | Starting cost | Best for |
|---|---|---|---|---|---|---|---|
| 1 | Hermes Agent | Nous Research | Your machine or server; Docker, Modal, Daytona backends | MEMORY.md, USER.md, SQLite FTS5 session search, self-written skills | Manual approvals, denied on timeout; hardline denylist | Free (MIT) + model tokens | Developers who want an agent that learns |
| 2 | OpenClaw | OpenClaw Foundation | Your machine or server; opt-in sandbox | Editable Markdown workspace files; SQLite sessions since 2.0 | Full host access unless sandbox is enabled | Free (MIT) + tokens; managed hosts ~$12–$49/mo | Widest chat-channel reach |
| 3 | Gemini Spark | Dedicated Google Cloud VMs | Google account context via connected apps | Asks before spending money or sending email | $19.99/mo AI Pro (US) | Google Workspace users | |
| 4 | OpenAI dots | OpenAI | OpenAI cloud computer; optional connected computer | Chats, ChatGPT memory and the dot’s own notes | Proactive research is read-only; passwords hidden from model | Included in eligible ChatGPT Pro and Business Premium plans | ChatGPT and Codex power users |
| 5 | Meta Muse | Meta | Muse Secure VM in Meta’s cloud | Learns preferences; forgets on request | Sentinel agent gates internet access; approvals for email and purchases | Free; $20 and $100 tiers | Everyday errands on a free tier |
| 6 | Perplexity Personal Computer | Perplexity | Your Mac plus Perplexity cloud | Not detailed publicly | On-device authorization, 2FA, audit trail | Pro or Max subscription | Mac users who need local files |
| 7 | NanoClaw, IronClaw, ZeroClaw | Open-source projects | Containers, WebAssembly or a small Rust binary | Varies by project | Isolation on by default | Free + tokens | Security-first self-hosters |
1. Hermes Agent (Nous Research)
What it is: Hermes Agent is an open-source agent from Nous Research.
It runs from a terminal UI, a native desktop app, a local web dashboard and messaging gateways. Parallel counts about 25 supported platforms as of October 2026, including Telegram, Slack, WhatsApp, Signal, Teams and email. Native iMessage arrived in June 2026.
Memory: Memory is the main reason to pick Hermes. It keeps persistent MEMORY.md and USER.md files and curates them with periodic nudges. Session history sits in SQLite with FTS5 full-text search and LLM summarization. Pluggable providers include Honcho user modeling plus 8+ community backends.
Hermes also writes its own skills. It turns completed complex tasks into reusable skills and refines them over time. Skills follow the agentskills.io standard. The /learn command captures a workflow as a slash command without hand-writing SKILL.md.
Security: The docs describe a 7-layer defense-in-depth model. Layers include user allowlists, DM pairing and human approval for destructive commands. Approval mode can be manual, smart (an auxiliary LLM rates risk) or off.
The defaults fail closed. An unanswered approval prompt is denied after 300 seconds. Your own deny rules survive even YOLO mode.
Nous is explicit about the limits. Its docs say the denylist does not sandbox a hostile or compromised agent. The project’s SECURITY.md names OS-level isolation as the one load-bearing boundary. For broad tool access, use a container backend such as Docker, Modal, Daytona or Vercel Sandbox.
Cost: The agent is free. You pay for model tokens from any provider: OpenRouter, OpenAI, Anthropic, or local runtimes like Ollama and vLLM. Nous Portal is optional, with Free, $20, $100 and $200 monthly tiers. Local models cut token spend to zero; see the one-click local model setup.
Comments: Hermes is the best open-source pick for most developers. It pairs the deepest memory stack with fail-closed defaults. Setup is still terminal-first, and the local backend offers no protection from a compromised agent.
2. OpenClaw (OpenClaw Foundation)
What it is: OpenClaw defined the category. It started in November 2025 as Clawdbot and went viral in January 2026.
OpenClaw 2.0 shipped as v2026.8.1 on August 31, 2026, built by 933 contributors. The latest release we covered is 2026.9.5, with atomic updates and plugin hot reload.
A central Gateway process routes messages, runs cron jobs and hosts a Control UI. OpenClaw reached 26 channels as of v2026.6, including WhatsApp, Telegram, iMessage, Teams, Matrix and Signal. Supported models include OpenAI, Anthropic, Google, DeepSeek, xAI, Kimi, Ollama and OpenRouter.
Memory: OpenClaw stores behavior in human-editable Markdown files. SOUL.md holds the persona, TOOLS.md the tool rules and USER.md what the agent knows about you. Version 2.0 moved session and transcript storage into SQLite for better persistence.
This design is transparent. You can open, diff and back up every memory file with ordinary tools.
Security: This is OpenClaw’s weak point. The default mode gives full host access, which WIRED called incredibly risky. An opt-in sandbox runs sessions in per-session Docker containers. It defaults to no network egress, a read-only root and all capabilities dropped.
The incident record is long. CVE-2026-25253 allowed one-click remote code execution via a WebSocket origin bypass. Visiting a malicious page was enough to steal auth tokens. It was patched quickly.
The skills registry was the bigger problem. Koi Security audited 2,857 ClawHub skills and found 341 malicious entries. 335 traced back to one campaign, now tracked as ClawHavoc. Antiy CERT later counted 1,184 malicious skills as the registry grew.
OpenClaw responded. It now scans every uploaded skill with VirusTotal and re-scans daily. Version 2.0 added protected credential requests and a team-scoped Secret Store. Treat ClawHub as untrusted code anyway.
Cost: OpenClaw is free under MIT. A survey of managed hosts found plans from about $12 to roughly $49 per month. Watch the heartbeat check. With expensive models, it can drive significant token spend.
Comments: Pick OpenClaw for channel breadth and the largest ecosystem. Turn on the sandbox, skip unvetted skills, and keep the gateway on loopback. For a deeper head-to-head, see our OpenClaw vs Hermes Agent comparison.
3. Gemini Spark (Google)
What it is: Google unveiled Spark at I/O on May 19, 2026. It runs on Gemini 3.5 Flash with Google’s Antigravity harness. Each user gets dedicated VMs in Google Cloud, so tasks continue with your devices off.
Spark works natively across Gmail, Calendar, Drive, Docs, Sheets, Slides, YouTube and Google Maps. MCP connections extend it to third-party apps. You can message it by text and email, and a macOS app shipped in June.
Memory: Spark’s memory is your Google account. It pulls context from the apps you connect through Personal Intelligence. Google has not documented a separate, editable memory store like Hermes or OpenClaw.
Security: You choose which services Spark can see. Google says it checks in before high-stakes actions such as spending money or sending email. Purchases run through Google’s Agent Payments Protocol, with merchant and transaction limits plus user approval.
Two caveats apply. The product page tells users to supervise closely and interrupt when needed. A leaked early onboarding screen warned that Spark may share your info or make purchases without asking.
There is one structural upside. Spark cannot directly access files on your local system. That limits the blast radius on your own machine.
Cost: Spark launched on the $99.99 AI Ultra plan. Since July 2026 it is included in the $20 US AI Pro plan. The Ultra global rollout excludes the EEA, Switzerland, the UK and Nigeria.
Comments: Spark is the easiest choice if your life already runs on Google Workspace. It is also the cheapest managed agent here. You trade portability: memory and actions stay inside Google’s ecosystem.
4. OpenAI dots (OpenAI)
What it is: OpenAI launched dots at DevDay on September 29, 2026. Each dot runs on GPT-6 Astra with its own cloud computer and browser. Plugins connect it to more than 4,000 apps.
You reach the same dot in ChatGPT, Slack, Microsoft Teams or a voice call. Texting is a limited beta for some US Pro users. For heavy work, a dot hands tasks to ChatGPT Work or Codex. OpenAI says the Codex harness powering dots is open source.
Memory: A dot draws on the conversation, relevant ChatGPT memory and its own saved notes. Those notes cover your preferences, decisions and ongoing work. Switching channels does not start a new dot or reset its memory.
Security: Between requests, a dot runs proactive research across connected apps. That research is read-only; follow-up actions need the right permissions. Your computer stays separate unless you connect it.
For website logins, dots use saved passwords without exposing them to the model. You can open the dot’s computer at any time and press Take over.
Cost: Your first dot is included. Chatting with it does not count toward ChatGPT limits, but its Codex and Work tasks do. Reports differ on the entry plan. BGR says dots require Pro 200 or Business Premium, not the $100 Pro tier. Check OpenAI’s plan page before you buy. The Pro rollout excludes the EEA, Switzerland and the UK.
Comments: Dots suit people who already live in ChatGPT and Codex. Memory carries cleanly across channels, and the read-only research default is sensible. It is the most expensive way into this category.
5. Meta Muse (Meta)
What it is: Meta launched Muse on September 8, 2026. It is powered by Meta’s Muse Spark model. You reach it through iOS and Android apps, muse.ai and WhatsApp. A Mac app followed on September 19.
Memory: Muse remembers what matters to you and makes unprompted suggestions. You can tell it to forget specific things it has learned.
Security: Muse has the most detailed published isolation design of the managed agents. It runs on Muse Secure VM, a dedicated machine that holds both the agent and your data. A separate Sentinel agent sits on the same VM, isolated at the system level.
Muse cannot reach the internet unless Sentinel approves. Sentinel asks you for permission when needed. Muse also asks before sending email or making a purchase, and keeps an audit trail.
Credentials go into secure storage that Muse can use without seeing. Checkout runs through Link by Stripe, which issues one-time-use cards. Meta says it plans a Muse Confidential VM encrypted with a key only you hold.
Cost: Muse has a free tier along with $20 and $100 monthly plans. It launched in the US, then expanded to Singapore and Australia.
Comments: Muse is the best free starting point for errands, shopping and scheduling. Its Sentinel design is a real architectural answer to prompt injection. Until the confidential VM ships, your data still sits on Meta’s cloud.
6. Perplexity Personal Computer (Perplexity)
What it is: Personal Computer merges Perplexity’s cloud agent with your local Mac. It needs macOS 14 Sonoma or later, and Perplexity recommends a Mac mini for 24/7 work. You can start and approve tasks from an iPhone.
The agent works with local files, native Mac apps and the web. Recent updates added hybrid compute that routes sensitive steps to a local model. Multiagent mode runs up to 16 agents that cross-check work.
Memory: Perplexity has not published a detailed memory model for Personal Computer.
Security: On Mac, a privacy gate and local model handle sensitive steps inside an isolated sandbox. Remote tasks on your desktop use 2FA. Perplexity also logs a full audit trail of agent actions.
Cost: Personal Computer requires a Pro or Max subscription. Rework lists Pro at $17/month and Max at $167/month on annual billing. Cloud work consumes credits.
Comments: This is the strongest option for Mac users who want an agent inside local files and apps. It has no chat-app channels, so you control it from Perplexity’s own apps.
7. Hardened OpenClaw alternatives: NanoClaw, IronClaw, ZeroClaw
OpenClaw’s security record spawned a wave of smaller, stricter forks. NeuralTrust’s October 2026 review shortlists Hermes, NanoClaw, ZeroClaw and IronClaw as the main open-source options.
- NanoClaw runs each agent group in its own Linux container that sees only mounted paths. In March 2026 it partnered with Docker to run agents inside MicroVM sandboxes.
- IronClaw is a Rust rewrite that runs untrusted tools in a WebAssembly sandbox. Secrets are injected at the host boundary.
- ZeroClaw is a minimal Rust binary. Composio says it runs on $10 hardware with under 5MB of RAM.
These projects trade ecosystem size for safety. Pick one when you handle regulated data or run tools you did not write. NVIDIA’s OpenShell sandbox work is also worth a look for hardening existing OpenClaw setups.
Also worth watching
- Claude (Anthropic): Rework lists folder-scoped access and approval before deletions, included in Claude Pro and higher plans.
- Manus: credit-based billing and an unusual 2026: it joined Meta, then resumed independent operations on September 1.
- Lindy: Team plan from $29.99/month for 3,000 credits; positioned as a work teammate.
- Poke: Free, $19 and $199 tiers, reached through Messages, WhatsApp and Telegram.
- Kimi Work: Moonshot AI’s local desktop agent, reportedly running a 300-subagent swarm.
Memory compared
The agents fall into 3 memory designs:
- Files you own: Hermes and OpenClaw store memory as Markdown on your disk. You can read, edit, diff and back it up.
- Vendor notes plus account context: dots and Muse keep their own notes and learned preferences. You can ask them to forget, but you cannot open the store directly.
- Account as memory: Spark relies on your Google data rather than a separate memory file.
If portability matters, choose files you own. Hermes and OpenClaw memory survives a model switch or a move to new hardware. Back up Hermes’s data directory before migrating machines.
Security compared
The key question is where the isolation boundary sits:
- Managed agents (Spark, dots, Muse) run on vendor VMs, away from your laptop. Muse adds a second agent that gates network access.
- Hermes puts approval prompts in front of risky commands and fails closed. Real isolation needs a container backend.
- OpenClaw gives full host access by default. Its sandbox is strong once enabled, but you must turn it on.
- Forks like NanoClaw and IronClaw make isolation the default.
Whichever you choose, apply 5 rules:
- Run self-hosted agents in a container, VM or dedicated machine.
- Grant read-only access first; add send and purchase rights later.
- Install skills only from sources you have reviewed.
- Keep approval gates on for email, payments and deletions.
- Review the audit trail weekly for the first month.
For isolation options, see our ranking of the best agent sandboxes.
Cost compared
Cost works differently across the 2 camps:
- Self-hosted (Hermes, OpenClaw, forks): the software is free. You pay model tokens plus hosting, or nothing beyond hardware if you run local models.
- Managed (Spark, dots, Muse, Perplexity): a flat subscription covers compute and the model, subject to usage limits.
Always-on agents burn tokens even when idle. Heartbeat checks, scheduled jobs and proactive research all call the model. On a frontier API, that background load can exceed your chat usage.
To control spend on self-hosted agents, route background work to a cheap or local model. Our ranking of open-source agent harnesses for local LLMs covers that setup.
Which one should you pick?
- You want your data on your disk and any model: Hermes Agent, on a container backend.
- You need WhatsApp, iMessage and the most integrations: OpenClaw, with the sandbox enabled.
- You live in Gmail and Google Docs: Gemini Spark on AI Pro.
- You already pay for ChatGPT Pro: OpenAI dots.
- You want a free tier for errands and shopping: Meta Muse.
- You work on a Mac with local files: Perplexity Personal Computer.
- You run untrusted tools or regulated data: NanoClaw or IronClaw.
FAQ
- What is the best personal AI agent in 2026? For developers, Hermes Agent offers the best mix of memory and safe defaults. For non-technical users, Gemini Spark and Meta Muse are the easiest to start with.
- Is OpenClaw safe to use? It can be, with hardening. Enable the sandbox, keep the gateway on loopback, and avoid unvetted ClawHub skills.
- Which personal AI agent is free? Hermes Agent, OpenClaw and the hardened forks are free software; you pay for model tokens. Meta Muse has a free tier.
- Do personal agents run when my computer is off? Managed agents do, because they run on cloud VMs. Self-hosted agents need an always-on machine, such as a home server or Mac mini.
Sources
- Hermes Agent security docs and work-machine guide
- Hermes Agent SECURITY.md
- State of Hermes Agent, November 2026
- Parallel: Hermes Agent vs OpenClaw
- Rework: Best Personal AI Agents in 2026
- eSecurity Planet: malicious ClawHub skills
- Immersive Labs: OpenClaw CVEs
- Meta: Introducing Muse
- OpenAI dots documentation
- Digital Trends: Gemini Spark on AI Pro
- TechCrunch: Perplexity Personal Computer on Mac
- NeuralTrust: OpenClaw alternatives

Asif Razzaq is the CEO of Marktechpost AI Media Inc.. As a visionary entrepreneur and engineer, Asif is committed to harnessing the potential of Artificial Intelligence for social good. His most recent endeavor is the launch of an Artificial Intelligence Media Platform, Marktechpost, which stands out for its in-depth coverage of machine learning and deep learning news that is both technically sound and easily understandable by a wide audience. The platform boasts of over 2 million monthly views, illustrating its popularity among audiences.
- Asif RazzaqMeet Thinking Machines Inkling 2 Mini: 26B Looped MoE Runs on a 32GB Mac
- Asif RazzaqBaseten Releases Bison-1 and Groundhog-1, Topping Decision Model Index
- Asif RazzaqRadixArk Ships DaisyRose-27B, a 27B Agentic Coding Model Trained With Reinforcement Learning Only
- Asif RazzaqNVIDIA AI Releases Nemotron-Labs-Sigma, a Continuous Diffusion Language Model with 14 checkpoints at 3B/8B