Pi vs Hermes vs Codex: Context Compaction and Memory

Three open-source coding harnesses, read out of their source: where each one cuts the conversation, what it writes down, and what it remembers tomorrow.

Every coding agent eventually runs out of context window. Pi, Hermes and Codex answer that in three different shapes, and this book reads all three answers out of the source rather than the documentation. Pi’s trigger, token accounting, cut-point rules, split turns and two summaries. Hermes’s three threshold corrections, free pruning pass, head/middle/tail with tail repair, micro-compaction and failure guards. Codex’s configurable limit, its choice of what to measure against it, and a summary written as a handoff to a different model. Then one debugging session traced end to end through every decision Pi makes, with arithmetic you can check line by line. Then memory: the file you maintain, the agent that forks itself after every turn to decide what was worth keeping, and the agent that writes only when you ask it to.

Read on your Kindle

We'll send this whole book straight to your Kindle — it opens natively, so you can resize the text, read fully offline, and it remembers where you left off. Nothing to download or manage.

Subscribe to send to Kindle

Sending to Kindle is for subscribers — subscribe to read the whole library on your Kindle.