long-horizon-agents

The long-horizon computer-use harness. Run AI agents across desktop apps and the CLI for extended periods while preserving task state and making reliable progress on complex workflows. Features fresh-context execution, durable verified state, independent auditing, recoverable progress, and native Claude Code / Codex / OpenClaw integration.

Long Horizon Terminal Benchmark with Dense Reward Grading

The local-first intelligence layer that gives AI agents durable continuity, explainable retrieval, portable context, and verified learning across harnesses.

Simple Long Horizon Agent - A simple yet effective AI agent for learning, experimentation, and long horizon work.

Test that your Claude Code skills, MCP servers, and CLIs actually work when an agent uses them — sandboxed YAML suites, activation checks, A/B experiments, CI gates.

An enhanced Codex CLI for complex, long-running software engineering tasks—resolving 89% more tasks than Codex at 27% lower cost, with up to 10× the effective context length.

Cortex: A Bidirectionally Aligned Embodied Agent Framework for Long-horizon Manipulation

A Plug-and-Play, Cost-Efficient, and Lightweight Memory Plugin for Long-Horizon LLM Agents

🔁 Build reliable recurring AI-agent systems: 956 resources, 22 operational patterns, 22 loop contracts, 8 runtime starters, an interactive atlas, and a structured dataset.

Official Repository for our paper: PlanBench-XL: Evaluating Long-Horizon Planning of LLM Tool-Use Agents in Large-Scale Tool Ecosystems

A workflow specification for autonomous agents

EvoX Genesis is an autonomous system for long-horizon software evolution that recursively builds, continues, and transforms complex software from high-level objectives.

Progressive disclosure retrieval for long-horizon AI agents over personal knowledge bases. Three-tier model: hot cache → FTS5 index → full read. Local-first, zero dependencies.

Recoverable long-horizon AI agents — a framework-agnostic reference harness + recovery-faithful live benchmark. Thesis: "Checkpoints Are Compactions" via Re-grounding Recovery. 0.x: v1.0 held until a powered live-LLM study confirms the claims.

Server-authoritative Minecraft AI agent with a Python LLM brain and Fabric + Carpet/Scarpet body. Autonomous navigation, gathering, crafting, combat, recovery, persistent memory, governed tools, and long-horizon play.

Self-hosted continuity and collaboration substrate for autonomous agents with bounded, recoverable memory and explicit restart orientation

Bounded context for long-horizon LLM agents: carry a small re-grounded signal instead of replaying the transcript, so a parent agent's footprint stays flat as the task grows. Recompute-verifiable; null results published; cheaper, not smarter.

A driver/worker/store discipline for bounded-context agentic execution: stateless workers, a born-tiered store, and structural audit independence so complex multi-session LLM work survives without losing the thread. Platform-neutral — the runtime ships as a Claude Code plugin.

Adaptive SDD harness for long-horizon coding agents. Lock product intent before agents write code.

DriftLedger turns your agent's stated plan into a versioned contract and reconciles it against live execution