Blog
-
Semantic WAL: Write-Ahead Logs for LLM Systems
2026-05-24
How to make probabilistic model behavior replayable: semantic WAL rows, deterministic reducers, event digests, and migration paths for prompt changes.
-
Cost-Based Context Assembly for RAG
2026-05-23
A retrieval planner should behave like a small optimizer: estimate marginal utility, enforce diversity, respect ACL scope, and spend tokens deliberately.
-
Effect Types for AI Agents
2026-05-22
Instead of giving an agent a flat tool list, classify tools by effect type and let the workflow engine enforce what the model may do.