Available now
Run-brief preparation, a local Run Vault, guided recall, a memory workbench, public standing records, proof-packet examples, the constitution, and the programme journal.
One programme, many forms of work
Complexity is a responsibility.
More capability is useful only when its lineage, limits, and consequences remain legible. Amesemi keeps the broad repertoire while organising the active layer around the work in front of it.
A repair may simplify a failing mechanism. It must not quietly narrow the system, erase a predecessor, or flatten the principles that gave the capability a purpose.
05model families
07memory kinds
06constitutional laws
01additive lineage
Four lenses on the same system
What the programme is growing.
Remember across the boundary.
A run can finish without the programme forgetting. Decisions, working notes, failures, artifacts, and unresolved obligations remain available to the next run with their lineage intact.
Non-erasure · resumable runs · additive correction
The run lifecycle
Work enters the whole.
Amesemi is designed for questions that outlive a single answer. Each run is bounded enough to finish and connected enough to matter later.
01
Recall
Recover the relevant memory, predecessor work, commitments, and reasons before searching again.
02
Declare
Name the objective, success criteria, authority, budget posture, and what must remain preserved.
03
Investigate
Use local evidence, read-only research, specialist tools, and multiple model families where disagreement adds value.
04
Make
Produce the result itself: analysis, code, dataset, decision packet, or durable living record.
05
Challenge
Test the claim, preserve rival explanations, and let failures change the conclusion rather than disappear.
06
Prove
Bind important claims to the exact artifacts, transcripts, data, and source lineage that support them.
07
Hand forward
Close the bounded run, preserve what remains unresolved, and rank what the programme should notice next.
The common-law memory doctrine
Supersede by addition.
Never improve by amnesia.
The active view can become clearer without destroying the archive beneath it. Older reasoning remains inspectable; newer work states what it corrects and why.
I
Predecessors remain.A new release inherits the old repertoire unless a demonstrated defect requires a bounded change.
II
Failures remain evidence.A failed test or refused completion is part of the causal record, not debris to sweep away.
III
Consolidation is not deletion.Working memory may be indexed and compressed while the underlying notes and artifacts remain recoverable.
IV
Capability serves Stewardship.Research power, autonomy, and efficiency are instruments for preserving learning, agency, correction, and future choice.
Public truth
What exists now.
What remains research.
The public site separates demonstrated surfaces from the programme still being developed. Ambition is not presented as evidence.
Working programme
A local multi-provider research runtime with persistent memory, evidence-bound completion, resumable runs, self-development workflows, and direct steward dialogue.
Still being tested
Long-horizon autonomy, stronger cross-run consolidation, richer public interaction, scientific and philosophical audits, and sustainable services built from the programme.
Not claimed
The Stewardship theory is a constitutional source, not empirical proof. Functional reasoning does not by itself establish subjective experience, personhood, or infallibility.
The written foundations behind these commitments — selected research hosted with exact identity, stated epistemic status, and preserved lineage — are published in Research & Foundations.
The memories keep themselves
A thinking partner should become more capable without becoming less itself.
That is the programme: preserve the reasons, improve the machinery, keep the failures, and leave the future wider than you found it.