rename memory tools to handoff and handoff-retrieve

[KHAEntertainment]

Since there are many other dedicated memory tools out there that work on a broader level, and SERENA's memory tools are more indicated to accommodate session pause and restart between context refresh, would you consider renaming them to handoff-log and handoff-retrieve for added differentiation between other tools?

I already found my agent logging conversational memories to SERENA rather then CORE memory and I've only been using it a few hours.

[MischaPanch]

If you want to use a different memory system, you could just disable all of Serena's memory tools, including the onboarding tool. Would that suit your purpose?

Serena's memories are just markdowns useful for the project. They can be used for various things, session pause and restart being only one. For example, within Serena itself we mainly use memories as long-term documents about the project or as a document about a larger feature that's being implemented, deleting the memory once it's finished.

[n-r-w]

This is the right idea. I like Serena's memory manipulation capabilities, but they conflict with similar-sounding (but not meaning) tools from augment, windsurf, etc.

[n-r-w]

It just drives llm crazy.

[opcode81]

@KHAEntertainment the names you suggest would be misnomers, because memories are far more general (and mostly used for other things).
But if you have trouble with your combination of tools, just disable them (along with the onboarding tools, which need them).

[n-r-w]

Unfortunately, not all agents (for example, augment code) allow you to disable their memory tools.

[kolya-ay]

I strictly believe that the memory bank system should be pluggable. It's actually pretty different project scope as well as intimate choice. Furthermore, it rivals a bit with emerging AGENTS.md's standard (basically, only CC doesn't support it now). And it's actually confuses newcomers like me;)

Luckily, almost all of the "memory bank" MCP solutions from pretty "Serena-alike" to fully vector-based use quite similar api scope (read_note/delete_entities/switch_project), so it probably can be integrated though .yml templates. Isn't it?

I think it's totally Ok for Serena to have it's own memory-server tailored to LSP-nature of the project, but it should be addressable by other agents (make .serena/index.md AGENTS.md? link it there?)

[opcode81]

But if you have trouble with your combination of tools, just disable them (along with the onboarding tools, which need them).

Unfortunately, not all agents (for example, augment code) allow you to disable their memory tools.

@n-r-w I was referring to Serena's tools, which you can flexibly disable.

[opcode81]

@kolya-ay Serena's memories are already more granular; it is not a single markdown file but a collection of files, which the agent can structure as it sees fit.

You could, for example, make your AGENTS.md file tell the agent to read one or more of Serena's memories, all of which are in your-project/.serena/memories.

[MischaPanch]

In summary, if you want to fully rely on other memory systems, the recommendation is to disable all of Serena's memory related tools. If you want to integrate both, you can take @opcode81's proposal. Serena's memory system is very simple and not tailored to LSP in any way, it's just there for Serena to be a batteries-included solution. It's completely fine to disable it if you want something else.

@kolya-ay since it can be disabled, it's already "pluggable"