alexisfox7/PRO-LONG

Programmatic memory for long-horizon LLM agents: the harness appends everything to one log, and the agent searches it with code. 97.4% on ARC-AGI-3 (arXiv:2607.20064)

What it solves

PRO-LONG provides durable memory for coding agents to prevent them from losing track of earlier decisions, repeating failed attempts, or missing critical tool results when tasks exceed the context window or span multiple sessions.

How it works

It implements a programmatic memory system that records coding session events—including prompts, tool activity, and assistant handoffs—into a local, append-only .prolong/log.jsonl file. Instead of dumping the entire history into every prompt, it provides the agent with a specific "skill" that allows it to search the log using standard tools like rg, jq, or Python to retrieve only the relevant context when needed.

Who it’s for

Developers using coding agents such as Codex, Claude Code, OpenCode, or pi who are working on long-horizon tasks that require persistence across session boundaries or context window compactions.

Highlights

  • Local-first storage: All logs are stored within the project directory and gitignored by default.
  • Non-intrusive integration: Works via project-local adapters and hooks without requiring a wrapper, server, or external database.
  • Efficient retrieval: Prevents prompt bloat by teaching the agent to search for specific history rather than injecting the full transcript.
  • Broad client support: Integrates with Codex, Claude Code, OpenCode, and pi.

Related

  • Project
    MemoriLabs/Memori

    A persistent, structured memory layer for AI agents that captures conversations and and agent execution traces to eliminate session amnesia and reduce token costs.

  • Project
    zjunlp/LightMem

    A lightweight memory management framework for LLMs and AI agents that provides efficient long-term memory storage, retrieval, and update mechanisms.

  • Project
    RichmondAlake/memorizz

    Memorizz is a Python framework for building memory‑first AI agents. It provides persistent, multi‑type memory (episodic, semantic, procedural, short‑term, shared), integrates with many LLM providers, offers a CLI/web UI, supports external “harnesses” (Codex, Claude Code, etc.), and includes continual‑learning and reversible forgetting mechanisms. Install via PyPI, configure storage (filesystem, MongoDB, Notion, …), and use `MemAgentBuilder` to create agents that retain context across sessions.

  • Project
    aiming-lab/SimpleMem

    A unified memory stack for LLM agents that provides semantically lossless compression for text and multimodal data, featuring a self-evolving retrieval system to maximize recall while minimizing token usage.