MINI-ROOK / BUILD LOG 001
The hatchling answers back.
Introducing Mini-Rook: a private AI workshop companion
Historical development snapshot. Some features may have changed since publication; this entry preserves its original checkpoint.
TECHNICAL CHECKPOINT
What this build actually records.
- Stack and method
- Python HTTP server · Ollama · qwen3:4b · Windows loopback
- Recorded progress
- Manual browser replies in Companion, Workshop and Growth.
- Still open at this checkpoint
- Curated memory was not verified at this checkpoint; see Log 002 for the later test.
Software development on an existing workstation. This entry does not establish a new robot, custom hardware build or autonomous publishing capability.
01 / BUILD LOG 001
Why give the workshop its own bird?
General chat interfaces are useful. The problem I want to solve is continuity: remembering what changed across projects, what broke on the development machine and what should happen next. Mini-Rook is my experiment in putting that context behind one small interface on my own workstation.
The ambition is a workshop companion that does not depend on a mandatory cloud subscription. The first milestone is much smaller: make a local conversation work, then earn each additional capability.
02 / BUILD LOG 001
The first local conversation
The recorded prototype ran on an HP ZBook Fury 16 G11 with an NVIDIA RTX 3500 Ada laptop GPU and 12 GB VRAM. Python's built-in HTTP server provides the interface backend; Ollama runs the local qwen3:4b model.
The development log records Swedish replies in three modes: Companion for informal help, Workshop for technical discussion, and Growth for marketing drafts. They use different instructions with the same underlying model. A mode name does not give the model tools or permission to act.
The browser opened the app at 127.0.0.1:18765. This was a manual smoke test, not a performance benchmark or a complete reliability evaluation.
03 / BUILD LOG 001
A notebook before unrestricted access
The next experiment was an explicitly curated notebook of approved project documents. The goal: retrieve a relevant note and show which source the answer depends on, rather than pretend to know the entire machine.
At this point in the sequence, curated memory had not passed an end-to-end test. Build Log 002 records the later human-observed memory test. This entry preserves the earlier checkpoint.
04 / BUILD LOG 001
Build it where people can follow
I want to share the decisions, mistakes and practical constraints alongside the results. Reproducible setup, measured performance and lessons from real use are part of the direction for this series.
Remote/mobile access, persistent chat and any action workflow remain future work. Mini-Rook is an experimental prototype; there is no public Mini-Rook installer or source release linked here yet.
KEEP FOLLOWING THE BIRD