Every month we spotlight the people who make Bernstein better. Here are April's highlights from the first month of public development.
What happened in April
Bernstein went from v1.0.0 to v1.8.8 in a few weeks. The pace was intense, and community contributions made a real difference:
- Architecture decomposition: 52 oversized modules broken into 22 focused sub-packages, each under 600 lines. The orchestrator monolith (4,198 lines) is now navigable, testable, and merge-conflict-free. Read the full story.
- 18 agent adapters: We started with 7 adapters and now support 18: Claude Code, Codex, Gemini CLI, OpenAI Agents SDK, Cursor, Aider, Amp, Kiro, Kilo, Qwen, Goose, Ollama, Cody, Continue, OpenCode, Cloudflare Agents, IaC, and a generic wrapper. Each adapter is a focused Python class under 200 lines.
- Cost-aware routing: The contextual bandit router learns which model handles each task type best. In our own runs, the bandit cut spend roughly in half compared to sending everything to the same model.
- Cloudflare cloud execution: Agents can now run on Cloudflare Workers with Durable Workflows, R2 artifact storage, and D1 state.
- Windows support: Full cross-platform compatibility contributed by @oldschoola: environment passthrough, Unicode safety, process management, and terminal handling.
Contributors
Thanks to everyone who contributed PRs, reported bugs, and tested edge cases this month:
- @oldschoola: Windows compatibility (3 merged PRs), codex config, task filtering, auto-PR
- @Ai-chan-0411: community spotlight template
- @alexanderxfgl-bit: spotlight generator script
-
@forfreedomforrich-eng:
--dry-runflag, trigger URL fix -
@TheCodingDragon0:
bernstein config diff, glossary - @internet-dot: HOL workflow
- @Beledarian: config path validation
All contributors are listed in CONTRIBUTORS.md.
How to get involved
Bernstein is Apache 2.0 and welcomes contributions of all sizes:
- Good first issues: curated tasks for newcomers
- Write a blog post: get published on bernstein.run
- Adopt an adapter: become the maintainer for your favorite agent
- Submit benchmarks: share your orchestration metrics
Top comments (0)