Main promise
A durable computer and workspace for people and their chosen agents.
A persistent personal assistant inside ChatGPT.
THE PERSONAL AGENT FIELD GUIDE
Dot, Gemini Spark, Grok Bot, Hermes Agent, Instinct, Muse, OpenClaw, Poke—and Matrix OS. Follow the differences that matter when the agent has real work to do.
READ THE TABLE
Each column summarizes public product material and Every's personal-agent comparison. The Matrix column describes today's product. A missing public claim is marked “not documented,” not treated as proof that a feature does not exist.
Compare Matrix with
| Feature | |||||||||
|---|---|---|---|---|---|---|---|---|---|
| At a glanceWhat kind of agent are you getting, and where does its work live? | |||||||||
| Main promise | A durable computer and workspace for people and their chosen agents. | A persistent personal assistant inside ChatGPT. | Ongoing tasks across Google's apps. | Named bots sharing a managed workspace. | An open agent you can run yourself or with Nous hosting. | An assistant centered on text messages and calls. | An assistant for goals and everyday life. | An open agent with self-host and managed-host choices. | A messaging assistant built around recipes. |
| Agent count | Multiple native agents and sessions on one computer. | One Dot at launch. | One Spark. | Multiple named Bots. | Multiple agents or profiles. | One agent. | One Muse. | Multiple agents. | One Poke. |
| Models | Choose installed coding agents; Hermes can use a configured provider. | GPT-6 Astra. | Gemini models; external models not described. | Cursor-managed models; user-supplied models not described. | Bring a provider or local model. | Instinct model; other providers not named. | Muse Spark; external models not described. | Bring a provider or local model. | Provider undisclosed; paid tiers cite frontier models. |
| Proactive work | Scheduled workflows and background sessions on always-on plans. | Research and scheduled work. | Scheduled and event-based tasks. | Schedules and triggers. | Schedules and triggers. | Proactive messages and calls; email/calendar triggers reported. | Suggestions based on goals. | Schedules and triggers. | Schedules and triggers. |
| Where it runs | Managed Matrix Cloud or manual self-host Preview. | OpenAI cloud; optional local computer access. | Google cloud. | Cursor cloud. | Your host or Nous hosting. | Cloud computer, per hands-on reporting. | Meta cloud. | Your host or a managed host. | Poke cloud. |
| Availability | EU and UK supported; German compute locations available. | Pro excludes EEA, UK and Switzerland at launch; other plan paths differ. | AI Pro/Ultra in supported rollout regions. | Trial and eligible paid plans. | Public download. | Invite-only beta. | Initial US rollout. | Public download. | Worldwide availability reported. |
| Price snapshot | Starter $20, Builder $100, Max $200/month, plus separate agent/model costs. | First Dot included on eligible Pro/Business Premium plans; check later usage. | AI Pro $19.99 or Ultra $99.99/month. | Cursor Pro $20, Pro+ $60, Ultra $200/month; extra usage may apply. | Software is free; hosting and models vary. | Free private beta; later pricing unpublished. | Free tier; Power $20, Maximum $100/month. | Software is free; hosting and models vary. | Free tier; Pro $19, Ultra $199/month; extra usage may apply. |
| Where you meet the agentThe same task feels different when you need to switch devices or channels. | |||||||||
| Desktop app | Matrix Desktop for macOS, Windows and Linux. | Documented. | Documented. | Documented. | Documented. | Not documented. | Documented. | Documented. | Not documented. |
| Mobile app | Native app in progress; phone browser works today. | Documented. | Documented. | Documented. | Not documented. | Not documented. | Documented. | Documented. | Not documented. |
| Web access | Open your existing computer in a browser. | Documented. | Documented. | Not documented. | Documented. | Not documented. | Documented. | Documented. | Not documented. |
| Messaging channels | Configured Hermes or OpenClaw channels; availability depends on setup. | Slack and Teams; limited US texting beta. | No direct message channel described. | Slack for Team Bots. | WhatsApp, Slack, Telegram and others after setup. | iMessage, WhatsApp and agent email. | WhatsApp. | WhatsApp, Slack, Telegram and others after setup. | Apple Messages, WhatsApp, Telegram and RCS. |
| Live voice | Depends on the installed agent and client; no universal Matrix voice claim. | In-app voice; no outgoing calls. | In-app voice. | In-app voice. | Desktop/CLI voice and Discord voice chat. | Phone calls, including proactive calls. | Voice announced; availability varies. | App/web Talk and optional phone plugin. | Not documented. |
| Sender identity | Your connected account or a configured bot identity. | Named Dot; no dedicated email at launch. | Your Google account. | Your account or Bot account. | Your account or bot account. | Agent email rollout; message identity unclear. | Your Meta account. | Your account or bot identity. | Your connected account or Poke email. |
| Systems and toolsThe practical surface an agent can use to turn a request into a result. | |||||||||
| Connected services | Supported integrations through granted connections. | Documented. | Documented. | Documented. | Documented. | Documented. | Documented. | Documented. | Documented. |
| Local files | Cloud workspace files; bring local files in deliberately. | With local computer access. | Mac beta. | Not documented. | Not documented. | Not documented. | Not documented. | Not documented. | Not documented. |
| Web browser | Browser on the Matrix computer. | Documented. | Documented. | Documented. | Documented. | Documented. | Documented. | Documented. | Not documented. |
| Computer use | The agent can work on its Matrix computer; inspect it yourself. | Cloud computer; optional local access. | Cloud plus local Mac beta. | Cloud with optional local commands. | Chosen host; optional Linux Bot Screen. | Cloud and phone, per hands-on reporting. | Cloud and local Mac app. | Chosen host. | Local commands through an unverified community bridge. |
| Terminal you can open | Yes, including named persistent sessions. | Not documented. | Not documented. | Documented. | Documented. | Not documented. | Not documented. | Documented. | Not documented. |
| Control and privacyPermissions, visibility and data controls matter more once agents can act. | |||||||||
| Approval for major actions | Set review boundaries; exact prompts depend on agent and connection. | Purchases and deletion prompts; Custom Rules can add boundaries. | Some actions prompt; stop or take control. | Auto Review; local commands prompt by default. | Configurable approval modes. | Messages and payments reportedly require approval. | Approval for email, purchases and sharing. | Configurable approval modes. | Not documented. |
| Connected access | Choose connections and their granted permissions. | ChatGPT app permissions; local access can be revoked. | Opt in to Google app access. | Plugin and team member permissions. | Toolsets, MCP servers and channel allowlists. | Linked apps can be granted or revoked. | Read/send scope and app disconnection. | Tools, channels and credentials are configurable. | Manage connected apps. |
| Activity visibility | Open files, browser, terminal and agent session output. | Activity View and inspectable cloud computer. | Task history and browser takeover. | Live computer view; Enterprise recording. | Session history; optional Linux Bot Screen. | Message thread; terms mention action records. | Planned and completed action log. | Logs on the chosen host. | Message thread. |
| Model-training control | Depends on chosen model provider and enterprise terms. | Personal opt-out; business data excluded by default. | Keep Activity required and data may train models. | Privacy Mode excludes customer data. | Depends on chosen provider. | Opt-out with safety-review exception. | Interaction-training opt-out. | Depends on chosen provider. | Maximum Privacy excludes training use. |
| Memory scope | Project files, instructions and saved conversations/sessions. | Own memory across channels. | Across chats. | Separate memory per Bot. | Across sessions within a profile. | Across chats. | Across chats. | Across sessions. | Across channels. |
| Memory controls | Edit saved project files and instructions; other controls depend on agent. | Correct in chat; reset removes Dot memories. | Turn off memory, delete chats or correct it in chat. | Correct in chat; direct editor not described. | Edit or delete local memory. | Delete indexed data; direct editor not described. | Ask Muse to forget. | Edit or delete memory files. | Not documented. |
| Delete agent data | Export or remove workspace files; account and backup deletion follow service terms. | Reset removes Dot, chats, memory and schedules. | Delete remote browser/computer data; chats separately. | Delete Bot and chats; clear shared computer separately. | Delete profile data; uninstall default agent for full removal. | Delete indexed data; disconnect apps separately. | Forget learned details; complete deletion not described. | Delete local state and revoke credentials separately. | Delete account; backups may persist briefly. |
What kind of agent are you getting, and where does its work live?
A durable computer and workspace for people and their chosen agents.
A persistent personal assistant inside ChatGPT.
Multiple native agents and sessions on one computer.
One Dot at launch.
Choose installed coding agents; Hermes can use a configured provider.
GPT-6 Astra.
Scheduled workflows and background sessions on always-on plans.
Research and scheduled work.
Managed Matrix Cloud or manual self-host Preview.
OpenAI cloud; optional local computer access.
EU and UK supported; German compute locations available.
Pro excludes EEA, UK and Switzerland at launch; other plan paths differ.
Starter $20, Builder $100, Max $200/month, plus separate agent/model costs.
First Dot included on eligible Pro/Business Premium plans; check later usage.
The same task feels different when you need to switch devices or channels.
Matrix Desktop for macOS, Windows and Linux.
Documented.
Native app in progress; phone browser works today.
Documented.
Open your existing computer in a browser.
Documented.
Configured Hermes or OpenClaw channels; availability depends on setup.
Slack and Teams; limited US texting beta.
Depends on the installed agent and client; no universal Matrix voice claim.
In-app voice; no outgoing calls.
Your connected account or a configured bot identity.
Named Dot; no dedicated email at launch.
The practical surface an agent can use to turn a request into a result.
Supported integrations through granted connections.
Documented.
Cloud workspace files; bring local files in deliberately.
With local computer access.
Browser on the Matrix computer.
Documented.
The agent can work on its Matrix computer; inspect it yourself.
Cloud computer; optional local access.
Yes, including named persistent sessions.
Not documented.
Permissions, visibility and data controls matter more once agents can act.
Set review boundaries; exact prompts depend on agent and connection.
Purchases and deletion prompts; Custom Rules can add boundaries.
Choose connections and their granted permissions.
ChatGPT app permissions; local access can be revoked.
Open files, browser, terminal and agent session output.
Activity View and inspectable cloud computer.
Depends on chosen model provider and enterprise terms.
Personal opt-out; business data excluded by default.
Project files, instructions and saved conversations/sessions.
Own memory across channels.
Edit saved project files and instructions; other controls depend on agent.
Correct in chat; reset removes Dot memories.
Export or remove workspace files; account and backup deletion follow service terms.
Reset removes Dot, chats, memory and schedules.
“Documented” means public product material describes a feature. “Not documented” means we could not confirm it from the cited comparison; it does not prove the feature is unavailable. Prices and eligibility are a snapshot and can change. Hermes Agent in this table is Nous Research's product, distinct from Matrix's Hermes runtime. See Every's original comparison.
Compare what each service promises with the actual work you need to review.