<?xml version="1.0" encoding="UTF-8"?><rss version="2.0"><channel><title>METABOLIQ Engineering</title><link>https://metaboliq.in/engineering/</link><description>Architecture decisions, experiments and lessons from building AI systems. Persistent memory, agent infrastructure and the foundations of intelligence.</description><language>en</language><item><title>A Backup Isn’t Recovery Until You’ve Restored It: Testing Memory, Embeddings and Access After a METABOLIQ Brain Restore</title><link>https://metaboliq.in/engineering/a-backup-isnt-recovery-until-youve-restored-it/</link><guid isPermaLink="true">https://metaboliq.in/engineering/a-backup-isnt-recovery-until-youve-restored-it/</guid><description>What an isolated Brain recovery drill verifies—and what checksum handling, file permissions, target guards and restored credential state still require us to explain.</description><pubDate>Tue, 17 Nov 2026 00:00:00 GMT</pubDate></item><item><title>A Memory’s History Should Survive Its Latest Edit: Revision Snapshots and Stale-Write Protection in METABOLIQ Brain</title><link>https://metaboliq.in/engineering/a-memorys-history-should-survive-its-latest-edit/</link><guid isPermaLink="true">https://metaboliq.in/engineering/a-memorys-history-should-survive-its-latest-edit/</guid><description>How Brain preserves earlier revisions, rejects stale updates and commits mutation evidence together—and where those guarantees stop.</description><pubDate>Mon, 16 Nov 2026 00:00:00 GMT</pubDate></item><item><title>A Model Name Isn’t a Reproducible Model: Pinning and Verifying Embedding Artifacts in METABOLIQ Brain</title><link>https://metaboliq.in/engineering/a-model-name-isnt-a-reproducible-model/</link><guid isPermaLink="true">https://metaboliq.in/engineering/a-model-name-isnt-a-reproducible-model/</guid><description>What pinned revisions, file checksums and an isolated offline inference test establish—and what loading, model identity and query caches still require us to qualify.</description><pubDate>Tue, 17 Nov 2026 00:00:00 GMT</pubDate></item><item><title>A Saved Memory Isn’t Necessarily Searchable Yet: Index Freshness, Failed Embeddings and Recovery in METABOLIQ Brain</title><link>https://metaboliq.in/engineering/a-saved-memory-isnt-necessarily-searchable-yet/</link><guid isPermaLink="true">https://metaboliq.in/engineering/a-saved-memory-isnt-necessarily-searchable-yet/</guid><description>What happens between saving a memory and semantic visibility: versioned indexing, failure recovery, cache behavior and the limits of concurrent search consistency.</description><pubDate>Tue, 17 Nov 2026 00:00:00 GMT</pubDate></item><item><title>An Audit Trail Should Explain an Agent’s Actions—Without Copying Its Secrets: Audit Events, Attribution and Privacy Boundaries in METABOLIQ Brain</title><link>https://metaboliq.in/engineering/an-audit-trail-should-explain-an-agents-actions/</link><guid isPermaLink="true">https://metaboliq.in/engineering/an-audit-trail-should-explain-an-agents-actions/</guid><description>How the accepted Brain foundation records selected agent operations, keeps memory content out of audit events, and exposes the limits of attribution and append-only storage.</description><pubDate>Mon, 16 Nov 2026 00:00:00 GMT</pubDate></item><item><title>Revoking an Agent’s Access Should Mean Something: Credential Expiry, Rotation and Authorization Boundaries in METABOLIQ Brain</title><link>https://metaboliq.in/engineering/revoking-an-agents-access-should-mean-something/</link><guid isPermaLink="true">https://metaboliq.in/engineering/revoking-an-agents-access-should-mean-something/</guid><description>What Brain checks when an agent presents a credential, what rotation preserves, and why ending future access cannot recall an already authenticated read.</description><pubDate>Mon, 16 Nov 2026 00:00:00 GMT</pubDate></item><item><title>We Built a Reranker—and Kept It Off: What METABOLIQ Brain’s Retrieval Evaluation Changed—and What It Didn’t</title><link>https://metaboliq.in/engineering/we-built-a-reranker-and-kept-it-off/</link><guid isPermaLink="true">https://metaboliq.in/engineering/we-built-a-reranker-and-kept-it-off/</guid><description>Why a local reranker remains experimental in METABOLIQ Brain: uncertain quality gains, query regressions, latency costs and the limits of relevance scores.</description><pubDate>Tue, 17 Nov 2026 00:00:00 GMT</pubDate></item><item><title>When an Agent Edits a Memory, Who Should Trust It?: Review States and Access Boundaries in METABOLIQ Brain</title><link>https://metaboliq.in/engineering/when-an-agent-edits-a-memory-who-should-trust-it/</link><guid isPermaLink="true">https://metaboliq.in/engineering/when-an-agent-edits-a-memory-who-should-trust-it/</guid><description>A concrete look at how Brain returns agent edits to review, restricts subsequent retrieval and preserves owner-visible history—and the limits of that design.</description><pubDate>Mon, 16 Nov 2026 00:00:00 GMT</pubDate></item><item><title>Why AI Agents Need a Shared Brain: Engineering Persistent Memory Beyond Context Windows</title><link>https://metaboliq.in/engineering/why-ai-agents-need-a-shared-brain/</link><guid isPermaLink="true">https://metaboliq.in/engineering/why-ai-agents-need-a-shared-brain/</guid><description>Why longer context windows alone cannot provide durable, governed knowledge for independent agents—and the memory architecture we are exploring at METABOLIQ.</description><pubDate>Mon, 16 Nov 2026 00:00:00 GMT</pubDate></item></channel></rss>