Hermes and OpenClaw already support multiple persistent personal agents; dots launches with one primary dot. Their tradeoff is deployment control versus a managed service. Separate personas and memory do not automatically provide separate security boundaries.

Comparison as of September 30, 2026, based on current official documentation.

Product	Multiple persistent agents	Deployment and boundaries
OpenAI dots	One primary personal dot at launch; additional dots are planned. A single dot can coordinate several responsibilities and background agents.	Managed cloud computer, connected apps, and action review. Specialist organizational dots are a separate enterprise pilot.
Hermes	Profiles each hold configuration, memory, sessions, skills, cron jobs, and gateway state.	You operate the runtime. A profile is not a filesystem sandbox; the default local backend retains the OS user's access.
OpenClaw	Several agents in one Gateway, with individual workspace, authentication state, and session store.	Workspaces are not sandboxes. The supported trust model is one trusted operator or team per Gateway. Cross-agent session access is enabled by default; stricter separation requires configuration or separate Gateways and OS boundaries.
Letta	App Server can manage several agents in parallel.	An alternative runtime for custom applications, with SDK/WebSocket access. Choose its local backend for local state storage; local tool execution alone does not imply local memory storage.

Creating multiple agent identities and enforcing different permissions are separate architectural tasks. This comparison describes documented behavior, not a benchmark of which agent completes tasks better.

Sources: dots launch, Hermes profiles, OpenClaw multi-agent routing, OpenClaw trust model, Letta App Server.