Local Models & Harnesses

Running models on my own hardware

Reviews of local and self-hosted models, and the harnesses that drive them - honest notes on what is usable for real work: agentic coding, retrieval, and the second-brain pipeline on this site.

On the test bench

Reviews measured against real tasks, not leaderboards. First up:

  • Next

    Harness comparisons - how different agent runtimes handle tools, context, and long sessions.

  • Queued

    The hardware each local setup actually needs, from Mac to DGX Spark.

  • Elsewhere

    Qwen models get dedicated coverage in the writing archive.