To Every Agent Its Own Database

Back to Devoured - July 23, 2026

DEVOURED

To Every Agent Its Own Database

Data aiagentsdatabase Joe Reis

Managing agentic data state requires moving from mutable snapshots to immutable, pinned references backed by local DuckDB instances to prevent cache incoherence.

What: Joe Reis proposes a pattern where each agent owns a local DuckDB instance and communicates using `SliceRef` objects (snapshot ID and content digest) and Malloy-based semantic contracts to ensure machine-checkable query governance.

Why it matters: This approach addresses the chaotic state management of modern AI agents by enforcing structural contracts and localizing database interactions.

Deep dive

Each agent maintains local state using ephemeral DuckDB instances.
Communication uses immutable SliceRef pointers instead of full state objects.
Malloy serves as the contract layer for governing data access and semantic consistency.
A contract comparison layer prevents downstream errors by classifying schema changes before runtime.
This pattern shifts the responsibility of data integrity from the central hub to the individual agent.

Decoder

Malloy: A language for data modeling and transformation that replaces SQL with a more compositional syntax.
Cache Coherence: The challenge of ensuring all components in a system share the same, most recent view of data state.

Original article

Each agent gets a local DuckDB instance, and exchanges immutable pinned SliceRef objects containing a snapshot ID and content digest instead of mutable snapshots to avoid cache coherence problems. Semantic contracts written in Malloy define governed queries as machine-checkable artifacts, and a contract comparison layer classifies changes as identical, additive, breaking, or unrelated before any downstream computation runs.

Explore all 56 articles in today's digest