Bub 0.5.0 is out.
- SDK for building your #Bub.
- sandbox: isolate the agent loop, tool calls, and context in different places.
- built-in traces.
- #Python CodeMode by @frostming90
thread ↓
Bub 0.4.4 is out.
- sidecar hook, with spill on it.
- one-line install with presets, gateway as a user service.
- Agent Plugins, portable skills and MCP drop in.
- FrontierHarness eval: thanks @guanlan for @runta credits, and @yihong0618 for @Kimi_Moonshot K3 API.
thread ↓
we ran it too.
Bub + K3, current master: 19/30.
as far as we know, the only #Python harness that finished the suite.
tiny hook-first runtime. come build your own #Bub .
A year ago the question was which model. Now it's which harness.
Pi, Exo, Claude Code, Codex, DeepSeek Harness and 4 others. Same model, same tasks, same runtime. 360 runs, 2 billion tokens.
Pass rates: 50% to 67%.
Cost per pass: $1.05 to $18.34.
Introducing FrontierHarness
A small thank-you to the open-source contributors who have been on this journey with us 💙
From Sep 13–16, quote this post with a PR that was merged before the campaign into any Bub, DeepChat, Lody, OOMOL, Proma, or Mem/Con repository to claim an Open Source Contributor Pack.
