Create

Sign in to ReadmeX

or

How we handle your data: Privacy Policy

ReadmeX

Your AI reading companion

What would you like to catch up on?

Pick a briefing, or ask your own question.

Your community and people briefings stay personal to you.

Story

SoduWork relay test uncovers macOS queue-blocking issue

AI summary

A hands-on test used SoduWork to switch one conversation among Pi 1.4.0, Qwen Code 0.21.8, and Google Harness² 0.2.0 while processing 120 cloud-gateway incident tickets. The test showed that shared artifacts and established rules carried across the handoff, but the first switch became blocked on macOS by runner-admission logic. After adding a temporary zero-concurrency bypass to a local runner, the Qwen Code and Google Harness² stages proceeded on the patched copy. Missing billable Qwen Code usage and a short test using one model provider prevented conclusions about total relay cost, long-history reliability, backend performance rankings, or provider failover.

In-depth insight

The report provides a working same-conversation, multi-backend workflow and a concrete diagnosis of the macOS queue blockage. It also shows that the temporary patch, incomplete billing data, short session, and single-provider setup limit broader conclusions.

用多个 Agent 写代码和跑数据时,我关心一个具体问题:换了工具,前面定好的规则和做完的文件能不能接着用?

在另一个工具里新开对话,通常还需要重新交代业务规则、字段口径和目录路径。按步骤选择不同底座或许能减少不必要的推演,但交接本身也要花时间,费用是否更低则需要另算。

开源的企业级 AI 原生引擎 SoduWork 把多个 AI 底座放进同一个界面,支持在本轮任务结束、满足切换条件后,在同一条对话里换底座。

我在 MacBook Pro 上接通了 Pi 1.4.0、Qwen Code 0.21.8、Google Harness² 0.2.0,拿一份 120 条云网关故障工单做了一轮“单会话三棒接力”,再查底层的对话 .jsonl 与 SQLite 用量表,核对交接结果和开销。

这轮接力始终使用同一条会话(01a1d12b-7ee4-71b4-9ea5-3e5667a1c7c4):Pi 先清洗并留下 R1~R4,Qwen Code 继续做区域评分卡,Google Harness² 最后尝试生成止血简报。中途的排队和失败重试也一并记在下面。

验证范围:这是本次短会话的结果,尚不能代表长历史交接也能保留全部细节。

#SoduWork#Pi#Qwen Code#Google Harness²#DeepSeek

Original · 15 min readRead the original
Group chat

No comments yet. Start the conversation.