An agent that forms cognition through experience
forme retains experience across ongoing interaction, builds an understanding of the user, itself, and the situation, and forms action intent within granted authority.
Facts enter context
Event / ProvenancePersistent experience
Experience stays in the next judgment
Events, provenance, time, relationships, and outcomes become traceable context. Later judgments can return to the same evidence instead of restarting from a single turn.
Facts are recorded before they are interpreted.
Inputs, actions, and environmental changes carry provenance and time so later judgment can cite them.
EventSourceTimeMemory becomes an individual cognitive model
The memory chain preserves what happened. The cognition chain forms a continuing model of the user, the agent, their partnership, and the situation. New judgments enter a candidate layer before evidence, conflict, and outcomes update the model.
A testable interpretation
New judgment stays revisable first.
Repeated experience and outcome evidence propose a candidate interpretation. One interaction cannot directly change the stable structure.CandidateUpdateProactivity
Proactivity follows a reviewable exit
Within an authorized observation scope, forme detects change, gaps, tensions, and commitments. Value, competence, and authority checks determine whether it internalizes, suggests, prepares, requests approval, or forms action intent.
Observe
Establish the current situation
Changes, gaps, tensions, and commitments enter the authorized observation scope.Change / Gap / Tension / CommitmentHarness runtime
Cognition reaches reality through the harness
The thought chain turns goals, situations, capabilities, completion criteria, and authority into action intent. The harness composes capabilities, checks authority, executes, verifies, recovers, and records events.
Policy ∧ Approval ∧ AutonomyEnvelopeOutcome loop
Outcomes change the next judgment
Outcomes, failure causes, and user feedback enter the evidence layer. They update memory, competence judgments, and candidate strategies, creating continuity across time.
Action intent enters reality.
The harness composes capabilities around the goal, policy, and authority envelope.
Action / PolicyCapabilities compose around the goal
Models, tools, skills, plugins, MCP, subagents, and executors form a replaceable capability layer. Goals and constraints determine the composition. Real invocation still passes through the harness.
Choose reasoning suited to the current judgment.
Models, subagents, and action patterns enter a ResourcePlan around the goal.
Go deeper
From one interaction to long-term continuity
Explore how the kernel governs action and how the cognitive system forms, revises, and preserves individual models.