Research Papers¶
The local research/ workspace contains two companion systems-paper drafts:
- Memorizz: A Memory-First Agent Harness for Efficient, Persistent Agentic Applications covers memory taxonomy and unit shapes, provider abstraction, prompt and semantic caching, context tokenomics, and harness boundaries.
- Memorizz: An Agent Memory and Continual Learning Platform covers immutable events, verified outcomes, bounded evidence, deterministic compilation, workflow-to-skill promotion, instruction authority, and governed forgetting.
The drafts are intentionally ignored by Git and are not part of the published
documentation or package. This protects the private authoring workspace; it
also means GitHub links to research/* are not valid until the authors choose
a separate publication location.
Build and open the local PDFs¶
From the repository root:
cd research
tectonic paper1-memory-first-agent-harness.tex
tectonic paper2-agent-memory-continual-learning-platform.tex
Tectonic writes each PDF into the current directory. It does not open a viewer automatically.
Underfull \\hbox messages are layout warnings, not build failures. A final
line such as Writing paper1-memory-first-agent-harness.pdf confirms that the
PDF was created.
Evidence and attribution¶
The platform paper credits the recency–importance–relevance retrieval mechanism in Generative Agents as inspiration for separating freshness, trust/utility, and relevance. Memorizz differs by keeping relevance as a recall signal and requiring reviewed, reversible retention decisions.
The drafts treat local benchmark runs as engineering evidence, not leaderboard
submissions. Results should retain their dataset revision, task count, provider,
model, scope, token/cost/latency method, official-grader status, and known
limitations. Do not publish local eval/results/ artifacts without reviewing
them for credentials, proprietary dataset content, tenant identifiers, and
unsupported statistical claims.