CoMind Core, at work.

Terminal recording · runs locally, no keys

v0.2 — continuity

The Mind That Wakes

An agent that wakes up already oriented — and stays itself.

Iris, a research assistant, wakes in the middle of ongoing work. You declare today's focus, and her whole pile of memory re-ranks around it. A conclusion she keeps using grows stronger. Then her learned memory is wiped — and when she wakes again, she is still the same agent, working toward the same goals.

• Wakes oriented: profile, goals, open threads, and promoted beliefs are assembled before any input.
• Re-ranks by intention: declare a focus and the whole pile reorders by what matters, not just what's similar.
• Reinforces through use: a conclusion the agent leans on is promoted and stays close across sessions.
• Clears without amnesia: clear() forgets learned memory but keeps identity, goals, and continuity intact.

Terminal recording · runs locally, no keys

v0.1 — the memory loop

The Detective Boundary Proof

If it isn't in the prompt, it has to be in the memory.

A robot detective works a case across several dispatches. Between each one, the conversation window is wiped clean — so anything the agent still knows had to come from CoMind, not from the context it was just handed. It's the clearest proof that recall is doing real work, selected by intent rather than dumped back in.

• The context window is wiped between dispatches — no hidden carryover in the prompt.
• Recall is weighed against the agent's standing lens: directives, values, goals, and current focus.
• A small, relevant set of memories comes back each turn instead of the whole history.
• demo:prove turns the same boundary into hard assertions that pass or fail.