Four personal agents, taken apart at the runtime level
Every comparison out there counts features. This one asks what the machine actually does after you press send — where the model runs, how a tool reaches it, and who can read your memory.
- 4runtimes taken apart
- 5axes, same for each
- 0scores published
- 2026-09-20last verified
Not yet scored
No numeric scores yet. Scores come from recorded task runs, and those runs have not happened. What follows is what we established about each runtime, which is the thing scores would have to rest on anyway.
| Axis | Pine | Instinct | Muse | Town |
|---|---|---|---|---|
| Where it runs | Server-side; browser work runs on your Mac | Server plus sandbox; leased cloud browser | One Linux VM per user | Server-side; native Mac reaches the system |
| How tools are exposed | 26names reach you, no schemas | 571by its own count; manual read enforced | 246permissions; schemas loaded per group | 1105listed; 177 offered to the assistant |
| Where memory lives | Server-side facts, no file to edit | Read-only markdown, log, secrets vault | Plain markdown in a home directory | Server-side wiki, memory and people |
| How you reach it | Web, Mac, outbound calls, REST API | WhatsApp, SMS, iMessage, its own email | Web, effectively only | Web and Mac; clock and email triggers |
| What we could not establish | Instruction text, call schema, carrier | Instruction body, host, model id | Everything sent to the model | Instruction set, personality placement |
Each column has a page of its own. Start with Pine, or read how we score first.