[{"data":1,"prerenderedAt":4},["ShallowReactive",2],{"readme:sol-pi":3},"\u003Cp align=\"center\">\n  \u003Cimg src=\"https:\u002F\u002Fraw.githubusercontent.com\u002FNVlabs\u002Fsol-pi\u002FHEAD\u002Fassets\u002Fsol-pi-hero.png\" width=\"100%\" alt=\"SoL-Pi: Scaling Auto-Research Loops for Efficient Agent Harnesses\" \u002F>\n\u003C\u002Fp>\u003Ch1>⚡ SoL-Pi: Scaling Auto-Research Loops for Efficient Agent Harnesses\u003C\u002Fh1>\n\u003Cp align=\"center\">\n  \u003Ca href=\"https:\u002F\u002Farxiv.org\u002Fabs\u002F2609.20519\" rel=\"nofollow ugc noopener\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FarXiv-2609.20519-B31B1B?logo=arxiv&logoColor=white\" alt=\"arXiv: 2609.20519\" \u002F>\u003C\u002Fa>\n  \u003Ca href=\"#getting-started\" rel=\"nofollow ugc noopener\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FGetting%20Started-Install-76B900\" alt=\"Getting Started\" \u002F>\u003C\u002Fa>\n  \u003Ca href=\"https:\u002F\u002Fgithub.com\u002FNVlabs\u002Fsol-pi\u002Fblob\u002FHEAD\u002Fdocs\u002Fconfiguration.md\" rel=\"nofollow ugc noopener\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FDocs-Configuration-555555\" alt=\"Configuration\" \u002F>\u003C\u002Fa>\n  \u003Ca href=\"https:\u002F\u002Fnvlabs.github.io\u002FSoL-Pi\u002F\" rel=\"nofollow ugc noopener\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FBlog-SoL--Pi-76B900\" alt=\"SoL-Pi Blog\" \u002F>\u003C\u002Fa>\n  \u003Ca href=\"https:\u002F\u002Fgithub.com\u002FNVlabs\u002Fsol-pi\u002Fblob\u002FHEAD\u002FLICENSE\" rel=\"nofollow ugc noopener\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FLicense-MIT-blue.svg\" alt=\"MIT License\" \u002F>\u003C\u002Fa>\n\u003C\u002Fp>\u003Cblockquote>\n\u003Cp>[!NOTE]\nThis repository contains the open-source version of SoL-Pi, a standalone extension for \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fearendil-works\u002Fpi\" rel=\"nofollow ugc noopener\">Pi\u003C\u002Fa>. It is not an official distribution of Pi.\u003C\u002Fp>\n\u003C\u002Fblockquote>\n\u003Ch2>💡 TL;DR\u003C\u002Fh2>\n\u003Cp>\u003Cstrong>Spend less without making the agent do less useful work.\u003C\u002Fstrong>\u003C\u002Fp>\n\u003Cp>SoL-Pi is a standalone extension for Pi that packages four reusable efficiency mechanisms discovered through scaled auto-research loops. It reduces repeated model turns, context replay, oversized observations, and unnecessary long-log reading while preserving the work and evidence an agent needs to finish a task.\u003C\u002Fp>\n\u003Cp>SoL-Pi installs on top of an unmodified Pi release. Every mechanism is opt-in and disabled by default.\u003C\u002Fp>\n\u003Ch2>Introduction\u003C\u002Fh2>\n\u003Cp>Long-running coding agents accumulate repeated work. A file edit is often followed by a predictable validation command. Large tool results are replayed long after their first use. Completed subtasks remain in active context, and a frontier model may spend a full request reading a log when only a few lines affect the next decision.\u003C\u002Fp>\n\u003Cp>SoL-Pi grew out of a broader question from our auto-research work: before scaling agent loops, can agents first make the harness itself more efficient? The search focused on constrained efficiency: reducing token traffic, inference work, and agent turns without stopping early, skipping verification, or hiding evidence.\u003C\u002Fp>\n\u003Cp>The standalone release contains four mechanisms that survived that process. They operate at different parts of the harness and compose through Pi's public extension APIs.\u003C\u002Fp>\n\u003Ch2>What SoL-Pi Adds\u003C\u002Fh2>\n\u003Ctable>\n\u003Cthead>\n\u003Ctr>\n\u003Cth>Area\u003C\u002Fth>\n\u003Cth>Mechanism\u003C\u002Fth>\n\u003Cth>What changes\u003C\u002Fth>\n\u003C\u002Ftr>\n\u003C\u002Fthead>\n\u003Ctbody>\u003Ctr>\n\u003Ctd>Tools\u003C\u002Ftd>\n\u003Ctd>\u003Cstrong>Action Fusion\u003C\u002Fstrong>\u003C\u002Ftd>\n\u003Ctd>An edit or write can run its follow-up validation command in the same tool call.\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd>Observations\u003C\u002Ftd>\n\u003Ctd>\u003Cstrong>ObservationPack\u003C\u002Fstrong>\u003C\u002Ftd>\n\u003Ctd>Repeated large text results become stable handles with exact paged recall.\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd>Delegation\u003C\u002Ftd>\n\u003Ctd>\u003Cstrong>Evidence-Preserving Reducer\u003C\u002Fstrong>\u003C\u002Ftd>\n\u003Ctd>Long diagnostic logs become compact receipts only when every retained quotation matches the archived source.\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd>Context\u003C\u002Ftd>\n\u003Ctd>\u003Cstrong>Online Context Compact\u003C\u002Fstrong>\u003C\u002Ftd>\n\u003Ctd>Completed plan steps become candidate points for Pi's native compaction, subject to economic and window-pressure checks; after a successful compaction, Pi continues the task in a new turn.\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003C\u002Ftbody>\u003C\u002Ftable>\n\u003Cp>The mechanisms share four rules:\u003C\u002Fp>\n\u003Cul>\n\u003Cli>\u003Cstrong>No Pi patches.\u003C\u002Fstrong> SoL-Pi imports public Pi APIs and does not vendor the Pi source tree.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Explicit opt-in.\u003C\u002Fstrong> A missing configuration leaves every mechanism disabled.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Preserve evidence.\u003C\u002Fstrong> Original observations remain available locally, and reducer failures leave the original result unchanged.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Use Pi's runtime choices.\u003C\u002Fstrong> Authentication, provider URLs, the main model, and shell behavior remain under Pi's control.\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Ch2>Technical Details and Core Insights\u003C\u002Fh2>\n\u003Cp>Read the \u003Ca href=\"https:\u002F\u002Fnvlabs.github.io\u002FSoL-Pi\u002F\" rel=\"nofollow ugc noopener\">SoL-Pi blog\u003C\u002Fa> for a deeper look at the technical details, design rationale, and core insights behind SoL-Pi, including how auto-research led to the four efficiency mechanisms and how they work.\u003C\u002Fp>\n\u003Ch2>Paper\u003C\u002Fh2>\n\u003Cp>Read our paper: [SoL-Pi: Recursively Scaling Auto-Research Loops for Efficient Agent Harness](\u003Ca href=\"https:\u002F\u002Farxiv.org\u002Fabs\u002F2609\" rel=\"nofollow ugc noopener\">https:\u002F\u002Farxiv.org\u002Fabs\u002F2609\u003C\u002Fa>.\u003C\u002Fp>\n",1789816652726]