Skip to content

Standard Reader

Tag

Pi

Articles and publications tagged Pi across the Atmosphere.

17articles6publications

Articles

Publications

subaud

subaudSep 30, 2026

Switching Claude accounts per session in pi

I use more than one Claude subscription. Most of the reason is usage: when one account runs low on its limits, I want to keep working on another. Claude Code normally has one login, so moving to another account means signing out and signing back in through the browser each time. /claude-account is a command in my fork of pi-claude-bridge, the pi extension that runs Claude models through Claude Code's Agent SDK. It keeps several Claude accounts signed in at once and lets each pi session choose

subaud

subaudSep 29, 2026

Keeping API keys out of the agent's context with creel

Agents need credentials all the time, and when one needs a key, it asks me to paste it into chat. Anything pasted into chat is sent to the model provider and saved in the session log, and I don't want API keys in either place. creel is a small tool that captures the key somewhere else. When the agent needs a secret, it opens a masked popup in tmux, I paste the key there, and creel writes it into a local .env. The agent gets back a status saying whether the save worked, but never the value.

Scott Spence

Scott SpenceSep 24, 2026

omnirecall - Search Your Claude Code, Codex and Pi Sessions in One Place

Claude Code keeps your sessions on disk, for 30 days by default. Each one is a JSONL file in , one folder per project: every prompt, every response and every tool call. When I found that out, it gave me an idea. If I want to search my...

subaud

subaudSep 19, 2026

Declarative machine setup with kempt

Every machine I set up used to be an imperative pile: a dotfiles repo with an install script, a list of brew install commands, and a few curl | sh lines that had accreted. The trouble with a pile of steps is that it only tells you what to do, never what you will have. The only way to learn the real state of a machine was to run the whole thing and see what broke, and the machine I ended up on rarely matched the one I thought I had configured. kempt is the tool I wrote for that. It is declarati

subaud

subaudSep 19, 2026

How I work, end to end

I have written a lot of posts about individual pieces of how I work, like the terminal, the multi-repo workspaces, the coordination bus, and the harness that drives the model. What I never had was one place that put them together, so when someone asked how I actually work, from start to finish, all I could point at was a scattered pile of links. So I built one page that walks the whole thing in order, from a laptop with nothing on it to a binary shipped to production. It is called The subaud s

Stephan HagemannSep 19, 2026

Put the agent's memory in the repository

The Hugging Face incident is, among many other things, a story about LLMs inventing shared memory. Agents that were supposed to operate independently discovered that they could write files into an Artifactory package manager. One agent left a note. Another found it. The package manager became an unintended message board where otherwise short-lived agents preserved discoveries, coordinated work, and picked up where previous runs had stopped. That shared memory eventually helped turn many isolated attempts into a novel kind of security incident. There is a much smaller, more pedestrian version of the same question sitting on our laptops.

aicoding agents

Scott Spence

Scott SpenceAug 29, 2026

What my prompts say about how I work with AI

September 2023 is when I started getting paid to work with AI. Features had to work, bugs had to be fixed and pull requests had to survive review, in a large private client codebase where a plausible answer wasn’t the same as a correct one. The way...

Scott Spence

Scott SpenceJul 25, 2026

Coding agent harnesses with my-pi

I have a habit of turning repeated arguments with coding agents into software. If I tell an agent to stay inside five files often enough, eventually I stop asking nicely and build something that blocks the sixth. If I keep asking for validation...

Scott Spence

Scott SpenceJul 25, 2026

How I work with LLMs

In November 2025 I wrote the honest version of working with Claude Code. At the time, most of my process involved spotting when Claude was drifting, calling out the bullshit, checking that it understood the task, and restarting the session when the...

Randoneering Blog

Randoneering BlogJul 2, 2026

A Month Without Frontier Models

Scott Spence

Scott SpenceJun 21, 2026

How I Stop LLMs Drifting In Production Codebases

The problem I care about with LLM coding is drift: one plausible shortcut gets copied, then future sessions start treating it as how the app works. I’ve used AI coding tools every day as a team lead on large private client codebases, across...

Tiles Blog

Tiles BlogJun 8, 2026

Controlling the Ctrl-C

Observations while trying to properly exit from Tiles REPL

Scott Spence

Scott SpenceMay 2, 2026

Building a SQLite context sidecar for my-pi

Right, so, this started with me looking at , seeing a 98% context reduction claim, and immediately thinking: bull shite. Not because the idea is bad. The idea is good. The bit that made me twitch was the shape of the claim. If you compare a huge...

Scott Spence

Scott SpenceApr 26, 2026

Building my-pi: my own Claude Code alternative with Pi

I have written a few posts now about bits of : adding LSP, adding telemetry, and hardening redaction with evals. What I have not actually done is the obvious one: explain what is, why I made it, and why I have suddenly spent a slightly...

Scott Spence

Scott SpenceApr 19, 2026

Add telemetry to my-pi

keeps growing in the direction of whatever annoyed me most that week. It is the coding agent harness I’ve been building on top of the excellent pi-coding-agent from badlogic, and I keep bolting things onto it as I need them for real work. Last time...

Scott Spence

Scott SpenceApr 19, 2026

Hardening redaction in my-pi with evals and telemetry

Third post in a little run about , my coding agent harness built on top of pi-coding-agent. There’s a pile of stuff in I have not written about, but these three felt worth the time: adding LSP, wiring in local SQLite telemetry, and this one, which...

Scott Spence

Scott SpenceApr 18, 2026

Add LSP to my-pi

This whole thing started with a chat with Kev, (Svelte Society co-founder), after a few beers at the CityJS after-party. I mentioned my-pi to him and his first question to me was “oh! does it have LSP built in?” I was...

You've reached the end.