Running a Compiler on a CDN
What happens when you run compiler infrastructure on a platform built for 50ms HTML rendering?
All Tags
Tag: technical
10 posts
Scaling the Number of Tools
The tool-count vs. reliability tradeoff only applies to static registration. Dynamic discovery changes everything.
Making Tools Discoverable
AI for AI: using models to predict how other models will search.
When There's Nothing to Embed
RAG assumes documents exist. For tool discovery, they don't—until you create them.
TypeScript Types for LLM Tools
LLMs understand TypeScript better than JSON Schema. Here's how we exploit that for better API execution.
Tools as Data, Not Code
Why retrieval-based AI tool use requires treating tools as data, not code—and what that makes possible.
Isolating Credentials
Most API specs model authentication as parameters instead of security schemes. We transform them so credentials stay out of LLM context.
Spec-Driven Execution
How we collapsed OpenAPI's encoding complexity into composable primitives.
Interface Synthesis
How we bridge the gap between what LLMs generate and what APIs require.
Extracting Operations
Why the operation—not the service—is the right unit for AI, and what it takes to extract one from a multi-megabyte spec.