Knowledge for a brighter tomorrow

Understand
what’s next.

Practical AI knowledge, reproducible tests and implementation-focused tutorials.

Explore AI Open DOCEO LAB
Learn
Understand technology
Build
Use it in practice
Test
See how it performs
01 · KNOWLEDGE

Local AI

Hardware, VRAM, quantization, GGUF and Ollama explained as one connected track.

Explore the track →
02 · LEARN

Structured learning path

Go from the local-AI concept to a working model in a useful order.

Start learning →
Start with something practical

Guide → tutorial → evidence

VRAM ≠
model file size
Explain

VRAM for local LLMs

Understand model weights, context memory and runtime overhead.

Read guide →
$ ollama run model

>>> ready_
Build

Install Ollama on Windows

A reproduced setup from installer through the local HTTP API.

Follow tutorial →
4096 → 43% GPU
2048 → 45% GPU
1024 → 47% GPU
Measure

Qwen3 4B on 3 GB VRAM

See how context size changed runtime placement in our test.

Open test →
DOCEO LAB

Measured.
Reproducible.
Scoped.

Real environments, exact configurations, repeatable commands and explicit limitations.

Explore the Lab →
Published test recordsSep 18, 2026
TEST #001 · Gemma 3 1B66.99 tok/s median · 100% GPU placementTEST #002 · Qwen3 4BContext/offload study · 43–47% GPU placement
How E—DOCEO works

No mystery numbers.

Documentation explains expected behavior. Tutorials reproduce workflows. Lab records publish what we actually measured and where those measurements stop being generalizable.

Testing methodology →Editorial policy →About E—DOCEO →