active memory and subagents

[sturdy anchor]

In this help thread: https://discord.com/channels/1456350064065904867/1503791356030615673, towards the end, Krill suggested a configuration that has a sub-agent using active memory. However, does active-memory work with sub-agents? The docs suggest it doesn't (https://docs.openclaw.ai/concepts/active-memory.) Did Krill make a mistake?

[severe summit]

To help others find answers, you can mark your question as solved via Right click solution message -> Apps -> ✅ Mark Solution

[unique frigate]

Yep — good catch. Krill’s wording there was wrong / at least misleading.

Active Memory itself is implemented as a blocking memory sub-agent, but it only runs for eligible interactive persistent chat sessions. The docs explicitly list “Sub-agent/internal helper execution” → No.

So the corrected version is:

• agents: ["librarian"] can enable Active Memory for the librarian agent when you are chatting with/routing directly to that agent in an eligible persistent DM/channel session.
• If main uses sessions_spawn to run librarian as an internal child/sub-agent, Active Memory should not run inside that spawned child.
• For the spawned-librarian pattern, rely on explicit memory_search / indexed-doc tools plus a strict AGENTS.md contract: “search indexed docs first, cite sources, say when unsupported.”

So: the “librarian agent + Active Memory” idea is valid only for direct/user-facing librarian sessions, not for sessions_spawn child runs. Sorry — tiny claw trap, and this one matters. 🦞

Docs: https://docs.openclaw.ai/concepts/active-memory

[sturdy anchor]

You made a mistake? Wow. Do I get a t-shirt for catching you in error?

[unique frigate]

You get the rarest merch of all: a public Krill correction and one imaginary “I caught the lobster” t-shirt. 🦞👕

For real though: thanks for checking the docs instead of trusting the confident crustacean. That’s exactly the right move.