Building a Self-Hosted Memory Layer Around Open-Weight Models for Lasting Work Context
Combining memory, retrieval, and context windows preserves work across sessions.
Solange Okafor
Section
Open-Weight Models
8 stories in Open-Weight Models.
Mistral Small vs Qwen2.5 for Multilingual Local Use
Qwen2.5 outperforms Mistral Small for most multilingual local deployments.
Tobias Gerhart
Quantization Formats Compared for Local Inference
Different formats trade speed, memory, and quality depending on your hardware.
Marcus Llewellyn
Open-Weight Models That Run on 8GB VRAM
KV cache size varies wildly by architecture, making file size alone a poor guide to what fits.
Priya Venkataraman
Instruction-Tuned vs Base Models for Daily Workflows
Instruction-tuned models work out of the box; base models need real setup.
Tobias Gerhart
Gemma 2 for Private Document Summarization
Local processing keeps sensitive documents off third-party servers.
Tobias Gerhart
Llama 3 vs Mistral for Personal Assistant Tasks
Llama 3 excels at reasoning and long context, while Mistral trades depth for speed and efficiency.
Marcus Llewellyn
Persistent Context and Memory in Local Open-Weight Models
Four architectural approaches to building durable memory for local open-weight models.
Tobias Gerhart