Blog
Thinking out loud.
Notes from building the context layer for AI agents. Engineering choices, architecture trade-offs, the failure modes we keep seeing in production agent stacks, and how we think about the problem space. Written by the team, unfiltered.
ENGINEERING2026-04-167 min read
The context-starvation problem in enterprise AI agents
Every enterprise AI agent project we've watched has failed for the same boring reason. Not because the models were weak — because they were context-starved.
ReadPRODUCT2026-04-145 min read
Why entity resolution is the missing primitive for AI agents
RAG solved retrieval. Nobody solved understanding. Entity resolution — collapsing the same record across systems into one node — is the part no connector does.
ReadARCHITECTURE2026-04-104 min read
Zero-copy vs ETL: why we don't move your data
Traditional data integration copies your data into a warehouse. mantle reads it in place. Here's why that architectural choice matters for AI agent infrastructure.
Read