Blog · Field Notes
Notes From the Memory Frontier.
Token autopsies, learning-loop design, benchmark reports. Numbers first, methods attached.
Latest
Posts.
2026-09-06The 8.59M-Token Session · 7 minWhat one observed coding session reveals about the re-reading tax, and how memory ends it.2026-09-18Best Local Coding Agent 2026 · 9 minOllama-first harnesses ranked by workflow: Anvaya, Cline, OpenCode, Aider, Continue. Privacy, models, memory, picked by constraint. Updated 2026-09 with Phase 1 measured data.2026-09-16We Benchmarked Six Agentic Coding CLIs: Phase 1 · 9 minRaw efficiency and OS footprint, measured: same models, same tasks, external verifiers, anv, jcode, codex, aider, claude and opencode, with every raw record published. Token economics, long-horizon accuracy, swarms, hallucination resistance and memory phases follow.2026-09-18Agentic Swarms, Explained Without The Hype · 8 minWhen a fleet of coding agents beats one, how supervision actually works, and the numbers we have (and have not) measured.2026-09-17Self-Healing CI With Coding Agents: A Practical Pattern · 7 minPlan-gate-apply, objective exit codes, monitors for long jobs, and memory that stops nightly retries re-learning the same failure.2026-09-16Why A Coding Agent Is Better Without A Runtime · 7 min12 ms to first frame, 4.2 MB idle, 25.6 MB median under load, what runtime-free buys, what it costs, and why we measured both.2026-09-15Agentic Coding vs Autocomplete: Different Tools, Different Contracts · 6 minOne predicts your next edit; the other owns a task to completion. The right choice depends on what failure costs.2026-09-18What YOLO Mode Should Never Do · 7 minAuto-approval is a decision about attention, not permission. The boundary lives in the sandbox, and there are jobs no amount of automation makes safe.2026-09-18Best Claude Code Alternatives 2026 · 9 minSeven alternatives ranked by what you are optimizing for (local execution, provider freedom, IDE depth, git purity) with the trade each one concedes.
Run Agents That Fit On Your Laptop.
25.6 MB median RSS. 25 agents ran in parallel on a Core 2 Duo with 4 GB RAM. Hundreds on your machine. Zero cloud required on the Ollama path.
Requires Rust/cargo to build from source. Linux and macOS today, Windows not yet supported. Pre-1.0, public beta. Pricing TBD.