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 and the Pi community have released Pi 1.0 and an experimental package called Pi Durable. The new framework is designed for agent applications that can run for long periods, use persistent storage and recover after interruptions; it does not replace the existing Pi coding agent.

Earendil and the Pi community have released Pi 1.0 alongside Pi Durable, an experimental framework for building agent applications designed to run for extended periods and recover from interruptions. Earendil says the package is separate from, and does not replace, its terminal-based Pi coding agent; the release gives developers another way to build agents that can use persistent storage and operate across different execution environments.

Earendil describes Pi Durable as a harness: software that stores conversations and coordinates model calls, tools and the environments where those tools run. Developers can configure agents and tools for individual conversations, while the framework manages their execution. The company says the package shares code and design principles with the Pi coding agent, including minimalism and malleability.

The release includes memory, SQLite and JSONL storage, as well as a conformance suite and benchmarks intended for developers implementing their own storage backends. Pi Durable also includes a Node execution environment for tools that need access to local files. Earendil says developers can implement alternative storage and execution interfaces, including setups where the harness and the tools run on different machines.

According to the report, a single process owns a storage backend at a time, with other clients attaching to that process. With SQLite, the harness keeps active transcripts, live tasks and pending submissions in memory while other data remains on disk until needed. Earendil says transcript compaction summarizes older messages before they exceed a model’s context window, allowing conversations with tens of thousands of messages to remain manageable in memory.

At a glance
announcementWhen: Announced alongside Pi 1.0; the source…
The developmentEarendil announced the experimental release of Pi Durable alongside Pi 1.0 as a framework for building persistent, long-running agent applications.

A Framework for Persistent Agents

Pi Durable addresses a limitation of conventional interactive coding-agent sessions: they are commonly tied to a running process and a person who can inspect a failure and restart the work. Earendil says its new framework is meant for agents that need to run for long periods, remain accessible through different surfaces and withstand process or infrastructure failures. If those capabilities work as described, developers could build applications whose agent sessions are not confined to a single terminal process.

The release also separates experimentation from the existing coding agent. Earendil says Pi Durable is a framework for agentic applications of different kinds, with coding agents among the possible uses. The company plans to apply lessons from those applications to the Pi coding agent when they prove useful, rather than making the coding agent itself depend on the new package immediately.

For developers, the project’s emphasis on small interfaces and a relatively compact codebase may make it easier to inspect or adapt than a larger, more prescriptive platform. Earendil estimates that the source code, excluding tests, is about 15,000 lines. That is the company’s description of the codebase, not independent evidence about ease of use, reliability or performance.

Amazon

SQLite database storage for developers

As an affiliate, we earn on qualifying purchases.

As an affiliate, we earn on qualifying purchases.

How Pi Durable Relates to Pi

Pi 1.0 is the release Earendil says reflects extensive hardening, maintenance and active development of its coding agent. That agent is designed to run on a local or remote machine, inside a terminal, with one person directing it. The company says that operating model remains unchanged and is what Pi 1.0 is intended to support.

Pi Durable is presented as a separate project for applications with different requirements, such as multiple people steering an agent or a conversation continuing across restarts. Earendil says it shares components with the coding agent, including pi-ai, while giving the team space to test other designs. The report says the package can run wherever a JavaScript runtime is available; its SQLite and JSONL storage code avoids Node APIs and can be adapted for environments such as Bun or Cloudflare Durable Objects.

The release report frames the package as an experimental starting point and invites developers to help develop it. The stated design goals include persistence, broad deployment options and conversations that can continue over time. Those goals describe the project’s intended capabilities; they do not establish that every deployment or failure scenario has been independently tested.

“Pi Durable was built specifically for long-running, durable, and malleable agents that can run anywhere.”

— Earendil, in its Pi Durable announcement

Amazon

persistent agent application frameworks

As an affiliate, we earn on qualifying purchases.

As an affiliate, we earn on qualifying purchases.

Reliability Claims Still Need Testing

Pi Durable is identified by Earendil as experimental, and the supplied report does not provide independent evaluations, production case studies or performance results. It is not clear how the framework behaves under specific combinations of crashes, storage failures, network interruptions or simultaneous user activity, nor what recovery guarantees developers can expect in each configuration.

The announcement describes the package as supporting agents that can run anywhere with a JavaScript runtime, but practical compatibility will depend on the storage backend, execution environment and deployment setup. The report also does not specify a general release timetable, a support policy, or which features are stable enough for production use. Those details should not be inferred from the stated design goals.

Amazon

long-running AI agent software

As an affiliate, we earn on qualifying purchases.

As an affiliate, we earn on qualifying purchases.

Developer Testing Will Shape Releases

Earendil is inviting the Pi community to help improve Pi Durable, so the next step described in the report is developer experimentation and feedback. Developers can assess the included storage options, build alternative backends against the conformance suite and adapt execution environments to their applications.

The company says findings from work on Pi Durable may feed back into the Pi coding agent when they prove valuable. The announcement does not name a date for a stable release or a specific follow-up milestone, so the package’s development schedule and the timing of any changes to the coding agent remain open.

Amazon

JSONL storage solutions for AI applications

As an affiliate, we earn on qualifying purchases.

As an affiliate, we earn on qualifying purchases.

Key Questions

What is Pi Durable?

Pi Durable is an experimental framework from Earendil for building agent applications. It coordinates conversations with language models, tools, storage and execution environments.

Does Pi Durable replace the Pi coding agent?

No. Earendil says the existing terminal-based Pi coding agent remains separate. Pi Durable is a broader framework that can be used to build coding agents as well as other agent applications.

What storage options are included?

The release report lists memory, SQLite and JSONL storage. It also describes a conformance suite and benchmarks for developers who want to implement their own storage backend.

Is Pi Durable a stable production release?

Earendil calls Pi Durable experimental. The announcement does not provide independent reliability testing, a production-readiness assessment or a timetable for a stable release.

Source: hn

EVERGREEN BESTSE

Evergreen bestsellers Picks

As an affiliate, we earn on qualifying purchases.

You May Also Like

Is GLM-5.3-Flash The Budget-Friendly AI Agent Solution We’ve Been Waiting For?

Z.ai releases GLM-5.3-Flash, a 320B parameter multimodal model with open weights and low-cost API, targeting AI agents needing long context and multimodality.

Use boring languages with LLMs

Experts suggest prioritizing consistent, less fragmented programming languages to improve the reliability of large language models’ code generation.

Claude Opus 5

Anthropic releases Claude Opus 5, an advanced AI language model with improved performance and safety features, marking a significant update in AI technology.

Android’s latest AI feature predicts what you’ll do next

Google is rolling out a new AI-driven feature for Android that predicts user actions based on habits and location, now available on Pixel 10 devices.