Pi 1.0
AIThis post was created with the assistance of artificial intelligence (AI).

TL;DR

Prime Big Deal Days · Oct 6–7Offer from Amazon

Get tech for your team delivered free — and shop member deals

  • Fast, free delivery on millions of items
  • Access to Prime Big Deal Days deals on October 6–7
  • Prime Video, Amazon Music and more included
Start your free Prime trial Free trial for eligible customers · Cancel anytime
As an affiliate, we earn on qualifying purchases.

Earendil has released Pi 1.0, a minimal, extensible agent harness with additions including Codemode, extension support for virtual models and deferred tool loading. The company also released Pi Durable as a separate experimental package for building longer-running agent applications; both releases are MIT licensed.

Earendil has released Pi 1.0, a hardened and extensible agent harness that adds Codemode, virtual-model extensions, deferred tool loading and other features. The company also released Pi Durable as a separate experimental package for building longer-running agent applications, extending Pi’s approach beyond coding agents and terminal use.

Earendil describes Pi 1.0 as the result of months of development informed by user issues and pull requests. The company says hundreds of thousands of people use Pi each week; that figure is the company’s own usage estimate, and the report does not provide a measurement period or methodology. Earendil says the release is intended to make Pi a stable tool while retaining its minimal design.

The listed additions include Codemode, which gives Pi native support for MCP and non-LLM models such as Jev and image models; extension support for virtual models; and deferred tool loading. The release also adds cache warming for Anthropic models, system messages that can change during a conversation while accounting for the transcript, a new terminal-interface theme, and full-screen mode as the default.

Earendil says Pi 1.0 is available through installation instructions for macOS, Linux and Windows on its site. Pi Durable is distributed separately as an experimental package through npm. The company says both releases are MIT licensed and points users to pi.dev for documentation and its GitHub repository for the code.

At a glance
announcementWhen: Announced today in the source report; b…
The developmentEarendil announced the release of Pi 1.0 alongside a separate experimental package, Pi Durable.

New Tools Keep Pi’s Minimal Design

Pi 1.0 formalizes a set of capabilities for people building and using AI agents, including ways to route work between models and extend the harness with virtual models. That can matter for workflows where different models handle different tasks, although the release announcement does not publish independent performance results or a comparison against other agent tools.

Pi Durable marks a separate product direction: Earendil says it is intended for longer-running applications and for access beyond the coding-agent and terminal setting. Keeping that work in an experimental package, rather than changing the core Pi product, lets the company test those uses while presenting Pi 1.0 as a stable release. How much adoption either release receives remains to be seen.

Amazon

AI agent development tools

As an affiliate, we earn on qualifying purchases.

As an affiliate, we earn on qualifying purchases.

From Agent Harness to Durable Applications

Earendil says Pi already worked with current models from major providers and was used as a daily coding agent and as a base for other agentic applications. The company presents the 1.0 release as a selection of features that survived a longer review process: it says it weighs each proposed addition against the complexity it would bring, rather than adopting every new development in agent tooling.

The announcement describes examples of the new capabilities, including a script that summarizes a week of commits and a virtual model that plans with Claude Opus, implements with GPT, and uses Jev to decide when to switch. These are examples supplied by Earendil, not independent evaluations. The company also points to a separate explanation of how Pi Durable was designed, while noting that the new package remains experimental.

“Pi does not work that way. We wait until something has proven itself, and only then do we consider adopting it; weighing its true functionality against its inherent added complexity.”

— Earendil, in its Pi 1.0 announcement

Amazon

virtual model extension software

As an affiliate, we earn on qualifying purchases.

As an affiliate, we earn on qualifying purchases.

Adoption and Durable’s Future Remain Open

The announcement does not specify how long Pi Durable will remain experimental, what milestones would lead to a stable release, or whether its package interfaces are expected to change. It also gives no independent usage data, detailed release notes, performance benchmarks or security assessment for Pi 1.0. The reported weekly user count comes from Earendil and is not accompanied by a definition of an active user or a measurement method.

It is also unclear which providers and models are supported through each new feature beyond the examples named in the announcement. The company describes Codemode and virtual-model workflows, but does not set out compatibility limits, costs or measured results. Users evaluating the tools will need to consult the documentation and code for implementation details.

Amazon

long-running AI application packages

As an affiliate, we earn on qualifying purchases.

As an affiliate, we earn on qualifying purchases.

User Feedback Will Shape Further Releases

Earendil says it wants feedback from users and plans to continue refining Pi in the open. Pi 1.0 can be installed using the instructions on pi.dev, and the company directs developers to its GitHub repository. Pi Durable is available now through npm as an experimental package.

The next visible developments will be further documentation, code changes and user reports about how the new features work in practice. Earendil has not announced a timetable for Pi Durable to leave experimental status or named a specific follow-up release date.

Amazon

terminal interface themes for AI tools

As an affiliate, we earn on qualifying purchases.

As an affiliate, we earn on qualifying purchases.

Key Questions

What is Pi 1.0?

Pi 1.0 is an agent harness that Earendil describes as hardened, minimal and extensible. The release adds tools for model routing, extensions, tool loading and terminal use.

What is Pi Durable?

Pi Durable is a separate experimental package for building longer-running agent applications. Earendil says it is meant to extend Pi’s approach beyond coding agents and terminal use.

What new features are included in Pi 1.0?

The announcement lists Codemode, virtual-model extensions, deferred tool loading, Anthropic cache warming, mid-conversation system messages, a new terminal theme and full-screen mode by default.

Are Pi 1.0 and Pi Durable open source?

Earendil says both are MIT licensed and links to the project code on GitHub. Pi Durable is marked experimental.

When can developers try the releases?

Both are available now, according to Earendil. Pi 1.0 has installation instructions on pi.dev; Pi Durable is offered as an npm package.

Source: hn

HALLOWEEN

Halloween Picks

As an affiliate, we earn on qualifying purchases.

You May Also Like

RHEO on Steam: One Toy, Every Screen

RHEO, the calming fluid art app, is launching on Steam, supporting PC, Steam Deck, Steam Machine, and VR, with seamless sync and shared experiences.

The Top Mesh WiFi Systems For Consistent Internet In 2026

Discover the best mesh WiFi systems of 2026, featuring WiFi 6 and WiFi 7 options for consistent, high-speed coverage across large homes.

Getting 50 GB/S Back From The Apple Neural Engine

Apple’s M3 Neural Engine reportedly reaches 50 GB/s throughput after addressing a performance issue, boosting AI processing capabilities significantly.

When AI Builds Itself: Inside Anthropic’s Evidence on Recursive Self-Improvement

Anthropic’s internal data shows AI systems like Claude are increasingly automating AI research tasks, raising questions about recursive self-improvement potential.