Living landscape
Agent Memory
Explore the products and frameworks with published evidence for agent memory capabilities.
How it works
Evidence before inference
This page is a published view of wiki evidence, not a popularity ranking or an exhaustive market directory.
- CollectPublished wiki entries provide the source material.
- ExtractDeterministic rules identify candidate entities and facts.
- Review with JEVJEV can review candidate relevance after deterministic extraction. It does not decide measurement credibility or invent business value; uncertainty fails closed.
- Canary and publishA versioned snapshot becomes visible only after evidence-integrity checks pass.
What inclusion means here
A product or framework appears only when published wiki evidence supports its inclusion. Each documented comparison capability links to the entry that described it. Vendor-described features are not hands-on verification; missing evidence stays missing.
Autonomous improvement loop
How Foreman improves the software
Foreman does not invent landscape evidence. It turns repeated, measurable product problems into bounded software work while the wiki and Spark ingestion continue independently.
- ObserveAnonymous aggregate latency and error signals reveal repeated user-facing problems. Prompts and personal data are excluded.
- DecideJEV may admit only reproducible, low-risk web or wiki-core work. Uncertainty, schema failure or protected paths stop the task.
- Build and reviewForeman gives the bounded task to an EVE implementation station, then a separate EVE reviewer checks the exact preview change.
- Verify or roll backTests, preview checks and production canaries must pass. A regression restores the previous deployment automatically.
Foreman may change only its allowlisted product paths. Evidence extraction, publication policy, workflows, database migrations and Spark maintenance remain protected and require the normal reviewed release path.