🧰 Harbor

The items that name it 6

• We ran AI agent swarms in Minecraft and Mindustry
  writeup2026-10-29

  Open-source framework behind Terminal-Bench, forked for multi-agent runs.

• [AINews] not much happened today
  article2026-10-03

  Framework underlying Terminal Bench; questioned for possibly letting agents modify trajectories before evaluation

• Welcome RL Environments to the hub
  article2026-09-28

  An environment framework that can load task directories from a Hugging Face Hub repository; its oracle agent runs a task's reference solution and the verifier scores the result without calling a model.

• Granite 4.2 LLMs: How They're Built
  article2026-08-25

  Agent harness (paired with Terminus-2) used to run the Terminal agent RL stage, driving multi-step command sequences in a live shell with rollouts spanning up to 64 environment turns.

• Aleph Alpha Kolibri-1 (78B MoE, Apache 2.0)
  repo

  Framework used for TerminalBench and SWE-Bench evaluations

• Terminal-Bench 4.0
  article

  Framework that Terminal-Bench 4.0 is built on.