What happened
Earendil released Pi 1.0 on October 1, 2026, with standard MCP support, Codemode, Deferred tool loading, virtual models, and a new Pi Durable package for long-running agents.
Why it matters
Pi 1.0 aims to keep the 'small harness' feel while adding features, so existing users get more capability without losing the simple design, and businesses get a stable version they can adopt with confidence.
What to watch
The test is whether Pi 1.0's added complexity — Codemode, Deferred tool loading, virtual models — stays balanced enough for users, and whether Pi Durable gathers enough feedback before any eventual mainline merge.
WHO IT HITSThis lands on developers and teams already using or evaluating open-source AI coding tools, who now get a stable Pi 1.0 with MCP support and multi-model flexibility, and on businesses that need a dependable, extensible coding assistant.
Summaries like this, in your inbox every morning.
Earendil's release of Pi 1.0 marks a maturation point for the open-source AI coding tool, which the company says is used by hundreds of thousands of people weekly. Over several months, Earendil incorporated user bug reports and code fixes to produce a stable version suitable for individuals and businesses. The update keeps Pi's long-standing 'small harness' philosophy while adding capabilities that were previously either absent or left to extensions.
The addition of standard MCP support lets AI interact with external tools like search and databases, but more tools mean more tool descriptions passed to the AI. Codemode addresses this by letting multiple tool calls run as JavaScript code, filtering results before they reach the model. Deferred tool loading further reduces the information passed by loading only needed tools on demand. Virtual models let users combine multiple AI models into one, as shown in an Earendil demo where Claude Opus planned, Jev classified, and GPT executed.
Alongside Pi 1.0, Earendil introduced Pi Durable, an experimental package for long-running agents that saves state so work can resume after a process ends. Earendil says it waits to adopt new features until they prove useful and keeps complexity in check. The outcome hinges on whether users find the new features genuinely helpful and whether Pi Durable's feedback loop leads to further integration.
Pick your industry and the AI tools you use, and get news related to your work every day.
Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →
Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.
