Feature request: per-agent auto-recall control

[itsmiso-ai]

Feature Request: per-agent autoRecall override control

Problem

Some local LLM backends output responses in a format that is not compatible with auto-recall (specifically: reasoning/thinking content is emitted to a separate field rather than the primary response field). A global `autoRecall: true/false` applies to all agents equally, making it impossible to enable auto-recall for compatible agents while disabling it for incompatible ones.

Proposed Solution

Allow per-agent override of the autoRecall setting via an exclude list or scope-based filtering.

Environment

• memory-lancedb-pro: 1.1.0-beta.10
• OpenClaw: 2026.4.1

[AliceLJY]

这个需求是合理的，当前确实只有全局 autoRecall: true/false，没有按 agent 做覆盖，所以混合后端场景会比较别扭。

先记两点结论：

• 现在没有 per-agent auto-recall 开关，临时 workaround 还是全局关掉后，对兼容 agent 手动用 memory_recall。
• 如果要做兼容性更好的增量设计，我更倾向先上一个最小版本，比如 autoRecallExcludeAgents / autoRecallIncludeAgents 这类 agent-id 级控制，而不是一开始就把 scope/filter/policy 全铺开。

如果你愿意提 PR，建议先按“最小可用配置 + 向后兼容默认值 + 测试覆盖 mixed-agent 场景”这个方向来。