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

TL;DR

Earendil and the Pi community released Pi 1.0 and an experimental package called Pi Durable. The package is a separate framework for building agent applications designed to run across environments and recover from process failures; it does not replace the existing Pi coding agent.

Earendil and the Pi community have shipped Pi 1.0 and an experimental package, Pi Durable, for developers building agent applications intended to run for long periods and recover from failures. Earendil describes the package as a separate harness for agents—not a replacement for the existing Pi coding agent—and is inviting community members to help develop it.

Buying for a business?Offer from Amazon

Get business pricing on tech for your team

  • Business-only prices and quantity discounts
  • Tax-exempt purchasing
  • Multiple users, one account, clear invoices
As an affiliate, we earn on qualifying purchases.

Earendil says Pi Durable is meant to support agents that can run in different environments, be accessed through different interfaces, and continue through extended conversations and failures. Its stated design also allows multiple people to steer the same agents. These are goals described by the project; the announcement does not provide independent testing or production reliability results.

The project defines a harness as storage plus the machinery for running conversations with language models, including tools and the environments in which those tools operate. Pi Durable ships memory, SQLite and JSONL storage options, along with a conformance suite and benchmarks intended for developers implementing their own storage backend. Earendil says the storage interface can be adapted to systems such as a key-value store or Postgres.

For tool execution, the package includes a Node execution environment that gives tools access to local files. Developers can implement other environments, including remote ones, so a harness and its tools can run on different machines. Earendil says the SQLite and JSONL storage code avoids Node APIs and can be adapted for Bun or a Cloudflare Durable Object. Those portability details are project statements, not claims of broad compatibility testing in the announcement.

At a glance
announcementWhen: Announced alongside Pi 1.0; the source…
The developmentEarendil released Pi Durable, an experimental framework for long-running agent applications, alongside Pi 1.0.

A Separate Framework for Persistent Agents

Pi Durable targets a different operating model from a coding agent run by one person in a terminal. In the existing Pi workflow, the user can inspect what happened after a process stops and ask the agent to continue. The new package is intended to make recovery and long-running operation part of the harness, with storage and execution environments that developers can adapt to their applications.

If its approach proves useful, developers could build agent applications that are less tied to one machine or one continuously running process. The design may also make it easier to connect different user interfaces or execution environments to an agent. However, the announcement presents an experimental developer framework, not a finished hosted service or a guarantee that agents will recover correctly in every failure scenario.

Products Worth Considering

Bose TV Speaker (2nd Gen) Compact Soundbar with Bluetooth and HDMI Connectivity

Bose TV Speaker (2nd Gen) Compact Soundbar with Bluetooth and HDMI Connectivity

As an affiliate, we earn on qualifying purchases.

How Pi Durable Relates to Pi 1.0

Pi 1.0 is the project’s release of its existing coding agent, which Earendil says is designed to run on a local or remote machine in a terminal under the direction of one person. The company says that agent remains focused on its current use case and is not being replaced by Pi Durable.

Instead, Pi Durable is a framework for building agentic applications, including coding agents. Earendil says it shares code with the Pi coding agent, including pi-ai, and the same stated principles of minimalism and malleability. The company also says lessons from applications built with Pi Durable may later inform the coding agent if they prove valuable.

Earendil describes the source code, excluding tests, as about 15,000 lines. It estimates that reading all of it would take about 150,000 tokens with GPT or 250,000 tokens with Claude, while noting that developers usually would not need to inspect the entire codebase. These are the company’s estimates and refer to source size and token counts, not runtime performance.

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

— Earendil, in its Pi Durable announcement

Reliability and Compatibility Still Need Testing

Pi Durable is explicitly described as experimental. The announcement outlines its architecture and intended behavior but does not report independent evaluations, reliability rates, or results from sustained production use. It also does not specify how the project will measure recovery after failures or how conflicting instructions from multiple human users will be handled.

The source says the framework can run wherever a JavaScript runtime is available and describes adaptations for particular storage and execution setups. It does not provide a complete compatibility list, deployment guidance for every platform, or evidence that every proposed configuration has been tested. The scope and timing of future releases are also not stated.

Community Testing Will Shape Development

Earendil is asking the Pi community to help improve Pi Durable. The next practical step is for developers to try the experimental framework, assess its supplied storage options and interfaces, and report where it fits—or falls short—for their applications. The included conformance suite and benchmarks are intended to help developers evaluate custom storage backends.

Earendil says ideas from projects built with Pi Durable may flow back into the Pi coding agent when they demonstrate value. The company has not announced a schedule for moving the package beyond its experimental status, so future milestones and support commitments remain unspecified.

Key Questions

What is Pi Durable?

Pi Durable is an experimental framework from Earendil for building agent applications, with storage and execution machinery for running model conversations and tools.

Does Pi Durable replace the Pi coding agent?

No. Earendil says the existing Pi coding agent remains focused on its terminal-based workflow. Pi Durable is a separate framework that can also be used to build coding agents.

What storage options does it include?

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

Is Pi Durable proven to recover from every failure?

No such guarantee or independent test result appears in the announcement. Crash recovery is a stated design goal, but the package is labeled experimental and its reliability in production remains to be established.

When will Pi Durable leave the experimental stage?

Earendil has not provided a timeline for a stable release or announced specific future milestones. It is asking the community to help develop and test the package.

Source: hn

You May Also Like

Elon Musk’s Vision For AI: How SpaceXAI Grok 4.7 Will Surpass All Others

Elon Musk announced that SpaceXAI’s upcoming Grok 4.7 AI model will outperform all current systems, though no benchmark data or release details are provided.

From Launchpad To Leaderboard: Grok 4.6 Elevates SpaceXAI To Fourth In AI Analysis

SpaceXAI’s Grok 4.6 reportedly surpasses Kimi K3 to reach fourth place on Artificial Analysis ranking, but verification details are pending.

How to Use AI for Entity Research the Smart Way

Ineffective entity research can be costly—discover how AI can transform your approach and unlock powerful insights you won’t want to miss.

Mistral Patent For “Code Implemented Tool Calls”

Mistral has filed a patent for a technology involving ‘code implemented tool calls,’ signaling potential advancements in AI tool integration.