SparklingKit: open-source local AI workbench runs six models at once on a DGX Spark
IanAndrewsDC · x · 2026-09-03
A user who spent $4,699 on a DGX Spark realized its 128GB unified memory could host a full team of specialized models simultaneously — not just a chatbot. That insight became SparklingKit, an open-source (Apache 2.0) self-hosted workbench with six tools in one panel: transcription with smart chunking (Markdown/JSON/SRT/VTT export), document OCR, dual-pane translation, visual grounding with up to 12 natural-language queries per image, and more.
The thesis: a gaming GPU forces you to pick one model, but 128GB lets you keep chat, ASR, translation, and image models all loaded at once. Everything runs locally via Docker with a one-line install, chaining results between tools with zero cloud dependency.
More from Infra
- Andy Masley: The data center backlash won't help AI safety — it may backfire — AndyMasley · 2026-09-03
- Claim: Without the datacenter buildout campaign, the US would be in a sharp recession — ZeeshanZiaML · 2026-09-03
- Together AI partners with Equinix and NVIDIA on a global distributed inference fabric — zhyncs42 · 2026-09-03
- Compute debate: multiplayer games already proved everyone rents cloud machines — kuza55 · 2026-09-03
- Moondream's Photon 2.1 adds streaming ASR, beats vLLM with up to 3.1x throughput — xeophon · 2026-09-03
- llama.cpp Metal gains M5 neural accelerator support, erasing MLX's prefill edge — MrPecunius · 2026-09-03