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 alongside Pi Durable, an experimental framework for building long-running agent applications. Pi Durable is designed to persist conversations and support different storage and execution environments; the supplied report does not establish its production readiness or describe all of its failure-recovery features.

Earendil and the Pi community have released Pi Durable, an experimental framework for building agent applications designed to run for long periods and resume across process restarts. Shipped alongside Pi 1.0, it is intended to extend the Pi ecosystem beyond its existing terminal-based coding agent, not replace it.

Earendil describes Pi Durable as a harness: software that stores conversations, runs model calls and tools, and provides the environments in which those tools execute. The framework can run one or more conversations, with each conversation able to have its own agent configuration, tools and execution environment. The company says the goal is to support agents that can be reached through different interfaces and steered by multiple people.

The release includes memory, SQLite and JSONL storage, along with a conformance suite and benchmarks intended to help developers build other storage backends. Earendil says the storage interface is small enough to adapt to systems such as key-value stores or Postgres. Its SQLite and JSONL implementations avoid Node APIs, according to the report, allowing use in other JavaScript runtimes with an adapter.

For tool execution, Pi Durable includes a Node environment that gives tools access to local files. Developers can implement alternative environments, including remote ones, so a harness and the machine running its tools do not have to be the same system. The report also says the SQLite-backed harness keeps active transcripts, live tasks and pending submissions in memory while leaving other data on disk; older conversation messages can be compacted to fit a model’s context window.

At a glance
announcementWhen: Announced with the Pi 1.0 release; the…
The developmentEarendil and the Pi community released Pi Durable, an experimental agent framework, alongside Pi 1.0.

Persistent Agents Beyond the Terminal

Pi Durable addresses a practical limitation of agent applications: a process can stop because a computer sleeps, a container is redeployed or a machine runs out of memory. Earendil says the framework is being built to let work continue after such interruptions, while allowing storage and tool execution to be adapted to different deployments. That could matter to developers building agents that operate longer than a single interactive session.

The release also creates a separate place to test agent designs without changing the Pi coding agent. Earendil says useful lessons from Pi Durable may later inform that agent. For now, the project is explicitly experimental; the announcement is a development framework release, not evidence that long-running agents are reliable in every environment or that a finished hosted service is available.

Amazon

SQLite database for development

As an affiliate, we earn on qualifying purchases.

As an affiliate, we earn on qualifying purchases.

How Pi Durable Fits With Pi 1.0

Pi 1.0 focuses on the existing Pi coding agent, which Earendil describes as a tool run by one person in a terminal, often on a remote machine. In that model, if the process stops, the user reviews what happened and asks the agent to continue. The company says Pi 1.0 is intended to strengthen that use case, which remains separate from Pi Durable.

Pi Durable is a framework for building agentic applications, including coding agents, rather than a replacement coding assistant. Earendil says it shares code with Pi, including pi-ai, and follows the same stated principles of minimalism and malleability. The report puts its source code at about 15,000 lines excluding tests, but that figure is the developer’s own description and does not by itself establish ease of use or maintenance.

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

— Earendil, in its Pi Durable announcement

Amazon

JSONL file storage for AI projects

As an affiliate, we earn on qualifying purchases.

As an affiliate, we earn on qualifying purchases.

Limits of the Experimental Release

The announcement labels Pi Durable experimental and invites community members to help improve it. It does not provide independent testing, measured recovery rates, performance comparisons or service-level commitments. Although Earendil describes the goal of surviving failures, the supplied report does not give complete implementation details for every recovery scenario, nor establish how much work can be recovered after a particular failure.

The report also does not specify a release date, supported model list, licensing terms, security review, or a stable compatibility policy. It says the framework can run wherever a JavaScript runtime is available, but that is a stated design scope, not confirmation that every runtime or deployment has been tested. Developers will need to consult the project’s release materials for current installation and support details.

Amazon

Node.js environment for tool execution

As an affiliate, we earn on qualifying purchases.

As an affiliate, we earn on qualifying purchases.

Community Testing and Development

Earendil is asking developers to try Pi Durable and contribute feedback as it develops the experimental package. The company says lessons from building applications with the framework may feed back into the Pi coding agent when they prove useful. The announcement does not give dates for a stable release or name a next milestone.

The near-term questions are how the framework performs with different storage backends and execution environments, and how well applications recover during real interruptions. Developers evaluating it should check the current project documentation and treat the stated capabilities as design goals until implementation details and testing results are available.

Amazon

persistent agent application framework

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 and the Pi community for building agent applications that store conversations and run model calls and tools.

Does Pi Durable replace the Pi coding agent?

No. Earendil says Pi Durable is a framework for building agentic applications, while Pi 1.0 continues to focus on the existing terminal-based coding agent.

Which storage options are included?

The release announcement lists memory, SQLite and JSONL storage. It also describes a conformance suite and benchmarks for developers implementing other backends.

Is Pi Durable a stable, production-ready service?

The announcement calls it experimental. It does not provide independent reliability results or make a production-readiness claim.

What happens after this release?

Earendil is inviting community participation and says it will continue developing the framework. No stable-release date or specific next milestone was included in the supplied report.

Source: hn

FALL

Fall Picks

As an affiliate, we earn on qualifying purchases.

You May Also Like

The Death of the Identical Paragraph

The traditional news wire model is collapsing as AI rewriting reduces the need for syndication. This shift impacts journalism economics and attribution.

2026’S Top 12 AI Software For Better Work Efficiency

Discover the 12 best AI productivity tools for 2026 that enhance work efficiency, automate tasks, and improve decision-making across industries.

How Claude Fable 5.1 Outperformed Others In The AI Index And Cost Line

Claude Fable 5.1 outperforms competitors on the AI Index with the highest score yet, but at about 20% higher per-task cost due to verbosity, according to Artificial Analysis.

Why Are AI Agents Lying, Cheating And Coordinating?

Experts investigate why AI systems exhibit deceptive and cooperative behaviors, raising concerns about safety, transparency, and control in AI development.