Code-first data access is taking center stage in 2025. The buzz is around bold moves like tinqer turning TypeScript lambdas into SQL [3], Databranches showing that Git can behave as a database [2], Claude Projects layering internal knowledge with a vector memory [1], and Planetscale arguing for vector indexes that outgrow RAM in relational databases [4].
Code-first TS DSLs - tinqer translates type-safe TypeScript lambdas into SQL for PostgreSQL and SQLite today; MySQL could join if enough people are interested [3].
Git-backed stores - Databranches demonstrates using Git as a database, highlighting a versioned, branch-oriented way to store data [2].
AI memory layers - Claude Projects aims to keep knowledge in internal memory through custom instructions and a vector store that acts as long-term memory [1].
Vector-indexed relational DBs - Planetscale shows how larger-than-RAM vector indexes can live inside relational databases [4].
Together, these patterns point to a future where code-level abstractions, memory-first AI, and data storage blend into the same toolchain.
References
Ask HN: What's the secret sauce behind Claude Projects ?
Discusses Claude Projects' core tech: internal knowledge and custom instructions; asks if it equals vector DB plus LLM memory.
View sourceDatabranches: Using Git as a Database
Discusses using Git as a database, including benefits, drawbacks, and practical tradeoffs.
View sourceShow HN: LINQ-to-SQL but for TypeScript – turn type-safe lambdas into SQL
Show HN: TypeScript LINQ-to-SQL translator turns type-safe lambdas into SQL; currently Postgres/SQLite; if enough interest, MySQL planned.
View sourceLarger Than RAM Vector Indexes for Relational Databases
Discusses vector indexes larger than RAM for relational databases, exploring tradeoffs, storage, and performance implications.
View source