Blog
Technical deep-dives, benchmarks, and insights from building high-performance AI infrastructure.
Notes from Shipping 28 Open-Source Projects in One Year
What we learned shipping 28 MIT-licensed projects under the neul-labs org — across agent runtimes, Rust accelerators, dev tools and governance. The themes that actually emerged, and the ones we expected but didn't.
Drop-in Compatibility as a Design Principle
Why most of the Neul Labs portfolio — rjest, rpytest, rninja, recurl, rewget, stout, fast-litellm, fast-langgraph, fast-crewai, fast-axolotl — refuses to change your config, your CLI, or your imports. The rule, the trade-offs, and the cases where we break it.
The Agentic Infrastructure Stack: What We're Building and Why
A reference architecture for an agentic stack — workspace runtime, memory, ORM-level policy, orchestration, payments — built from the Neul Labs primitives: agentvfs, brat, fastagentic, ormai, mcp-pay and memorg. With pointers to each repo and the open problems we still see.
Why We Chose Rust to Accelerate Python AI Infrastructure
Python's GIL and single-threaded ORMs create bottlenecks in production AI agents. Rust + PyO3 delivers 3x-700x faster drop-in replacements for LiteLLM, LangGraph, and CrewAI with zero code changes.
How to Make LangGraph Checkpointing 700x Faster with Rust
LangGraph checkpoint serialization is the #1 performance bottleneck in production agent deployments. fast-langgraph uses Rust to achieve 737x faster serialization, 151x faster deserialization, and 2.8x faster end-to-end execution.
How to Run Multiple AI Coding Agents in Parallel with brat
brat is an open-source multi-agent orchestration harness that coordinates Claude Code, Aider, Codex, and other AI coding tools on shared codebases with crash-safe state, merge queues, and a real-time dashboard.
Python Task Queue Without Redis: How FastWorker Replaces Celery + Redis
FastWorker is a brokerless Python task queue that eliminates Redis and RabbitMQ. It supports priority queues, auto-discovery, distributed workers, and a built-in monitoring dashboard — all without an external message broker.