HACKHackers
in the Loop
Join community →

Projects / Completed

Finished and closed.

Completed builds and research live here alongside projects that were retired or folded into other work. Each entry states how it ended.

Updated November 2, 2026.

All projects →Active projects →

Retired

Magnesium

One place to run and watch coding agents across a small fleet of machines: a web app for desktop and phone, backed by a server on each machine that ran agents. It reached its first goal, Claude Code and Codex threads with supervised approvals, then grew too big and unwieldy and was retired in October 2026. The work moved to fxbot, which starts small and grows only by add-ons.

Explore Magnesium on GitHub (archived) ↗
Project lead
Shannon Duncan ↗
Hardware & resources
  • Agent machines across the Lab

Became tensorless

P100 engine

Custom kernels to run Qwen3.8-27B, a modern 27-billion-parameter model, on one Tesla P100. The engine is now part of tensorless, which ships it with tuned models and a one-command launcher.

Explore tensorless on GitHub ↗
Project lead
Shannon Duncan ↗
Hardware & resources
  • Tesla P100
  • DGX Spark pair for the research agents

Became tensorless

P100 PN3/PN4 quantization

Quantization formats built for the P100, made from Qwen3.8-27B's original weights and checked against a full-precision baseline so smaller never quietly means worse. PN3 is now tensorless's first model.

Explore tensorless on GitHub ↗
Project lead
Shannon Duncan ↗
Hardware & resources
  • DGX Spark
  • Tesla P100

Retired

Agent Blade

A clusterable board for swarms of agents. Autonomous agents and very small local models ran on ESP32 prototypes; the first prototype put 6 ESP32-S3 and 3 ESP32-P4 chips on one PCB. Later experiments tried small SoCs, low-power GPUs behind the blades, and older Android phones as inference accelerators. The project is retired.

Project lead
Shannon Duncan ↗
Hardware & resources
  • Microbench
  • First prototype: 6 × ESP32-S3 + 3 × ESP32-P4 on one PCB

Completed

Cluster monitoring & HUD

Metrics, logs and dashboards for the whole Lab, with status screens on the rack.

Project lead
Shannon Duncan ↗
Hardware & resources
  • Lab controller
  • Rack status displays

Completed

Cluster provisioning

Repeatable setup for the Lab's machines, virtual machines and Kubernetes clusters, with the isolation needed to run other people's work safely.

Project lead
Shannon Duncan ↗
Hardware & resources
  • Cluster network, VM hosts and attached hardware

Completed

LiteLLM gateway

One gateway in front of every model the Lab serves, from the DGX Sparks and the GPU server, alongside cloud models. Each key sees only the models it is allowed to use.

Project lead
Shannon Duncan ↗
Hardware & resources
  • Cluster model-serving resources

Completed research

8 GB model testing

Tested a quantized Ternary Bonsai 2 27B on memory-constrained hardware: whether it fits, how fast it decodes, and how good its output stays.

Model files ↗Model announcement ↗
Project lead
Shannon Duncan ↗
Hardware & resources
  • BC-160
  • GPU bench

Completed build

Updated GPU Bench

A 4U GPU server now holds eight data-centre GPUs, four V100s and four P100s, with 192 GB of GPU memory between them. It is the Lab's main GPU machine, and its P100s are Thermite's research lanes.

Project lead
Shannon Duncan ↗
Hardware & resources
  • 4U GPU server
  • 4 × V100 32 GB
  • 4 × P100 16 GB

Abandoned

ZD Web Server

Serving the ZD harness and editor over the web, with a headless way to work with agent projects across the cluster. Development stopped.

Project lead
Shannon Duncan ↗
Hardware & resources
  • N150

Abandoned

fxClaw

An agent system in the vein of OpenClaw, Hermes, PicoClaw and NanoClaw, built in Zig with fx.sh as its foundation. Development stopped.

Project lead
Shannon Duncan ↗
Hardware & resources
  • MS-A2

Completed

pod-pi

Named, persistent Pi coding workspaces in rootless Podman containers. They can connect to a configured model provider, including models hosted on the DGX Sparks; pod-pi does not host a model server.

Explore pod-pi on GitHub ↗
Project lead
Shannon Duncan ↗
Hardware & resources
  • Linux hosts
  • Configured model provider, including DGX Sparks

Completed prototype

esp32-fx

Runs the fx coding agent on native Linux on an ESP32-S3. WAMR runs the fx WebAssembly core, and a Seeed reTerminal E1001 e-paper display shows the agent’s response. Model inference uses a remote provider.

Explore the esp32-fx source on GitHub ↗
Project lead
Shannon Duncan ↗
Hardware & resources
  • Microbench
  • Seeed reTerminal E1001 with ESP32-S3

Completed research

DeepSeek Flash 4.1

The research phase using a pair of DGX Sparks is complete. Performance results have not been supplied for publication here.

Project lead
Shannon Duncan ↗
Hardware & resources
  • 2 × DGX Spark

Completed build

GPU cooling shrouds

Cooling shrouds for the Lab's P100, P4, and V100 cards are complete. Installation and measured cooling results have not been documented here.

Project lead
Shannon Duncan ↗
Hardware & resources
  • P100, P4, and V100 GPU cards

Completed research

ESP32 Needle 3 decode research

A follow-on phase explored faster decode while retaining eight layers and checking output quality. The work is complete; performance findings have not been supplied for publication here.

Explore the ESP32 Needle 3 repository ↗
Project lead
Shannon Duncan ↗
Hardware & resources
  • Microbench
  • pod-pi
  • pi-autoresearch

Abandoned

microWASM

Development stopped. Existing WebAssembly runtimes such as Wasmer, Wasmtime, nodewasm, and wasmCloud are better options for this work.

Project lead
Shannon Duncan ↗
Hardware & resources
  • MS-A2

Abandoned

threaded

Development stopped in favor of Google's agent-substrate project.

Project lead
Shannon Duncan ↗
Hardware & resources
  • MS-A2

Merged into another project

An agent on your wrist

The standalone wrist-agent idea was folded into A tiny pet on your wrist, which continues as an active project led by Shannon's daughter.

Project lead
Shannon Duncan ↗
Hardware & resources
  • K15
  • ESP32-S3 wearable

Good company for your curiosity

Bring what you’re working on.
Or what you’re wondering about.

You don’t need a project or Lab access to take part.

Join community →