Skip to content

Public engineering catalogue

Small tools.
Hard boundaries.

Neul Labs builds inspectable components for performance-critical AI infrastructure, agent systems and the developer workflows around them.

03 engineering families Source-linked projects Version-specific claims

Three families, and how to tell which one you need

The catalogue is organised by the problem a tool solves rather than by the language it is written in. Performance Accelerators sit inside a Python process you already have and make one part of it faster. Agent Infrastructure sits around a running agent and controls what it can coordinate, schedule and read. Developer Tools sit in the loop a human or an agent runs dozens of times a day. A team usually adopts from one family for months before it touches another, and nothing here assumes you take more than one.

What is going wrongFamilyWhere to look first
One CPU core pinned while an agent service queues requestsPerformance AcceleratorsAccelerators
Checkpoint or serialisation cost dominates a multi-step graphPerformance Acceleratorsfast-langgraph vs LangGraph
Several coding agents conflicting on one repositoryAgent InfrastructureAgent Infrastructure
An agent needs production data and nobody has written the rules downAgent Infrastructureormai
The edit-test loop is slow enough that people stop running itDeveloper ToolsDeveloper Tools
Background jobs need a broker the service does not justifyDeveloper Toolsfastworker

What a catalogue entry commits to

Every project listed here has a marketing site on its own subdomain, reference documentation under docs.neullabs.com, and a source repository. Where a source URL did not resolve at the last audit, the entry says so rather than quietly linking somewhere else. Licences are recorded per repository and per release; there is no blanket portfolio licence claim, and the repository metadata is the authority.

Performance figures on these pages are workload- and version-specific. They were measured on the hardware and inputs described alongside each project, and the linked method is the claim — not the headline multiplier. The reasoning behind that policy is set out in the engineering evidence method.

Frequently asked questions

Do I need to adopt a whole family?

No. Each project installs on its own and none of them requires a sibling to be useful. The families are a way of navigating the catalogue, not a bundle.

Are these products or open-source projects?

They are open-source projects. Neul Labs separately offers scoped engineering work, described on the solutions page, and the two are deliberately kept apart: nothing here requires a commercial relationship.

How do I evaluate one honestly?

Reproduce the benchmark on your own workload before you believe the number. Each repository documents its method, and the comparison pages name the cases where the alternative is the better choice rather than pretending there are none.

What if the tool I need is not here?

The full index at all projects is longer than these three families, and the 2026 landscape groups every entry by category with its current status.

01

Engineering family

Performance Accelerators

Rust-powered drop-in replacements that make your existing AI frameworks production-ready. No code changes required.
Explore the family ↗
02

Engineering family

Agent Infrastructure

Production-grade orchestration, workflow automation, and data access for AI agents at any scale.
Explore the family ↗
03

Engineering family

Developer Tools

Faster testing, faster builds, faster everything. Rust-powered replacements for everyday developer tools.
Explore the family ↗