How We Teach Machines to Think, Without Asking Them to Be Human
Ape Space might look simple on its surface. But you know what they say about calm waters.
Simplicity requires engineering violence: cutting away everything that slows down, and building only what accelerates.
Welcome to the machinery.
APE2 — The Little Engine That Could
Why We Are Building This
Large language models are astonishing… and also extremely dumb in a couple of important ways:
They have no memory.
No continuity.
No thinking.
Just single-turn magic tricks.
Every sense of “coherent thought” is an illusion created by:
- duct-taped prompts
- endless context stuffing
- fragile agent frameworks
- and hope
So we started to design the future: an AI system, that doesn't try to be human, but that makes humans better at what they are already good at: intent, taste, choice, intuition – the Promethean spark. But make it a scalable fusion reactor for creative thinking! We looked at existing agent architectures, and we quickly realized, that what was out there didn't match our ambitions.
What we saw:
- conceptual baggage
- over-engineered SDKs
- frameworks trying to be everything and becoming nothing
- enough spaghetti to feed Italy
What we wanted:
- small and simple
- clean and consistent
- fast and parallel
- predictable, evaluatable and deterministic
- dev-friendly
So instead of compromising, we did the only reasonable thing:
We built our own agent engine.
From First Steps, To Thinking At Machine Speed
The first version was literally a Python class wrapped around Burr.
Functional.
Cute.
Like duct-taping a rocket to a bicycle.
Then we started to work on APE2 — forged from the same idea but built for cognitive computation at scale.
APE2 is:
- Python at its cleanest
- Pydantic at full throttle
- deterministic execution
- fully schema-enforced LLM calls
- parallel everything
- designed for maximal cognitive throughput
- provider-agnostic with pluggable adapters, auditing and eval
We have thrown thousands of thinking pipelines at it simultaneously.
APE2 just shrugged and kept going.
Parallel.map, parallel.join, loop, branch, human-in-the-loop, multi-stage pipelines — all smooth, all stable.
APE2 is our philosophy in code:
An exoskeleton for your mind.
Thinking, at machine speed.
ADL — The Language Of Thought (In YAML Disguise)
Every AI framework wants you to write code.
We wanted something better:
a declaration of how thinking should unfold.
Meet ADL, our Agent Definition Language.
ADL gives us:
- native parallelism
- native tool calling
- native human gates
- structured cognition
- clean thinking state transitions and recalls
- zero spaghetti
The entire cognitive flow:
described in a few lines.
It’s not public yet — we’re still refining the spec — but it already powers every agent in Ape Space, from Spark, to Writer, to the experimental pipelines we haven’t told anyone about (oops).
ADL is the blueprint;
APE2 is the engine that turns the blueprint into motion.
What This All Means For You
If you have a creative, unscalable problem – we want to hear from you. We are curiously exploring new ways of applying our technology to problems that seemed unsolvable before.
Let’s build something impossible together – reach out: team@apesonfire.com