Comparisons
Comparisons investigate architectural questions across selected systems. They contribute to design-space exploration by identifying alternatives, examining combinations and bounding claims about constraints and consequences. They are not product rankings. Follow the citations to the frozen analysis records to check a conclusion's scope. An unrepresented combination remains open unless evidence rules it out.
Agent memory
- How memory reaches the next decision — who initiates retrieval, how selection works, and which consumer receives the result. Contrasts Hermes, Nemsy, OpenKB, Hindsight and sqz; distinguishes catalog visibility, recoverable content and model-visible bodies from reliance.
- How corrections propagate through retained memory — store changes, index freshness, prose repair and stale consumer copies across Graphiti, Hindsight, Mem0, OpenKB, SQLite Graph Memory and Project Wiki.
- What the current corpus shows about agent memory mechanisms
— a frozen landscape of 243 analyses, dated 2026-11-09, at inputs commit
bf2da04db. Covers admission, read-back, curation and trace learning, with evidence filters and withheld conclusions. - Memory comparison table and CSV — find candidate implementations and inspect their scoped evidence. See the dataset boundary before interpreting counts.
- Current per-system memory accounts — build-time listing, which may be newer than the landscape and memory-table snapshots.
The Agent Memory topic maps the research questions these materials address.
Further research topics
Runtime & Control, Knowledge & Verification and Workflow & Recovery currently provide questions and per-system reading paths. The future-work agenda prioritizes interactions and decision-critical uncertainties, rather than coverage for its own sake. Cross-system articles can compare alternatives where inspected evidence supports them; experiments are needed where the question concerns actual behavior or benefit. No equivalent structured datasets exist for these lenses yet.
Interpret and reproduce
Methods & Data explains the analysis method and evidence levels. Comparison inputs and regeneration is the operational reference for population selection, profile versions, evidence interpretation and rebuilding the memory outputs.
Research Topics · Systems Directory
Complete file listing (generated at build time)