Latest posts
Measured, reproducible notes on local AI agents — every claim with code you can run. Proof, not vibes.
- Local AI
Meet Evarness: The Bench Gets a Name, and Its First Capability Graduates
The harness bench this series has been building now has a product name — Evarness — and a public home. But the real announcement is a discipline: nothing moves to the open-source repo wholesale. Capabilities graduate one at a time, each one finished, tested, and introduced by a post like this one. The first graduate is the assurance spine: canonical traces, invariant contracts, proof bundles, offline verification. One digest, byte-identical across five environments, is the receipt.
- Local AI
The Trace Is the UI: Evarness Writes Its First Post, and the Interface It Argues For Is a File
Evarness now has a blog of its own, and it lives where a product's claims should live: inside the docs site, subject to the same build gates as the docs — a post cannot cite a digest the build didn't just reproduce. Its first post makes the design argument behind the render artifacts: an assurance tool's most valuable interface is portable evidence, not an app. One self-contained HTML file carries the graph, the replay, the verdicts — and in the proof browser's case, the entire proof bundle, extractable and verifiable offline. This is the personal-side pointer: why I built it that way, what to click, and what the page deliberately refuses to claim about itself.
- Local AI
Simulation First, Then Go Live: I Built a Bench That Proves an AI Harness Before It Touches Real Data
Building a local, privacy-first agent taught me one discipline worth writing down: never let an AI harness meet real data until its declared invariants have been verified — reproducibly, in simulation, where every tool and inbox is a safe scripted stand-in. I built Harness Lab to make that discipline mechanical: prove the harness in simulation first, then go live by rebinding one tool. This post is the method, the five rules the bench enforces, and the proof — my agent's email path run both ways, the two traces the same shape event for event, differing exactly where going live must differ. The proof is one command — and the machinery behind it has just found a public home.
- Local AI
The Spine, Drawn: I Made My Agent's Governance a Graph You Can Sweep — and Can't Bypass
Last post I argued the harness is a deterministic spine and the guards, not the model-judges, hold the line. This one earns the next claim: I rebuilt that spine as an explicit graph on a bench, swept every governance knob, and watched the safety line appear as a number. When governance is topology instead of an aspect, the bypass isn't a test you keep passing — it's an edge that doesn't exist.
- Local AI
The Harness Has a Spine: Why Deterministic-First Beats Pure Agentic for Local AI
Two experiment campaigns on a local, privacy-first agent — a tuning-lever sweep and a judges-on red-team. Both times, reliability and safety came from the deterministic scaffolding, not the model or the knobs around it. For local agents, the leverage is a harness with a spine.
- Local AI
Local AI Agents in 2026: The Model Isn't the Bottleneck, the Harness Is
An eagle-eye view from three experiments on local open-weight models, run on an Apple-Silicon laptop with MLX. The model on your machine is already good enough for most real work; what decides whether it works is the engineering around it — and every failure I found was a locatable, fixable harness problem, not a model limit.
- Local AI
The Buffer-Size Cliff: The One Setting That Stops Your Local AI Agent Hallucinating
A sharp quality cliff at buffer_size ≈ 0.5 × prompt_tokens makes local AI agents stop calling their tools and start hallucinating — measured on six open-weight instruct models on Apple Silicon.