§ 1 · THE AUTHOR · ALEXANDRE NOTO

Manence came from a need.

I'm an entrepreneur and I've been working with AI intensively for three years. Manence is the set of rules I eventually wrote down so that work would hold over time.

Portrait of Alexandre Noto

I founded Intuition Software, which has been building and publishing the Jobaffinity ATS since 2009; I am a partner and its director of innovation.

I also run the Alex Déclic channel and the declic.media outlet, where I decode AI for anyone who wants to understand that world.

Three years of use taught me one thing: with agents, on projects that last, things drift. The first few weeks, everything moves fast. Then I re-explain context I've already given twice, a mistake fixed a month earlier comes back, a three-week-old draft ends up looking like a decision. It's not that the AI lacks memory: it has one. It's that everything piles up and nothing gets put away.

Manence is the fruit of that experience and of my research, tested in real conditions: it runs first at Intuition Software, on the ordinary work of a company. A rule that doesn't survive a real week of work gets cut from the framework.

Manence is open source because I think it can help a lot of people who work the way I do. And because this framework was built on ideas that others shared freely: giving back is the least I can do. Feel free to reach out through the contact page.

The system exists because I needed it.

§ 2 · LINEAGE

Manence is built on an inheritance.

Many concepts helped me think Manence through. Here are the ones that had the most impact on the construction of this project. They are documented in more detail in the repo.

Andrej Karpathy
Software 3.0: the model is the processor, context is the RAM, scarce and expensive. The whole thesis of the framework follows from it.
JP Noto
Living Reference / Sliding Canon. Seven of his mechanisms are in the framework, carried over in two waves — details and sources in source record 06.
OKF Google
The convention for writing knowledge that an AI can read as well as you can.
CoALA
The taxonomy of agent memory, which separates what is of record from what passes.
Anthropic
Context engineering, and the canonical building blocks of an agent that works.
OpenClaw
The identity conventions of an open source project.