Real documentation with "how", not "what"

[simonachmueller]

That seems like a nice addition to the opencode. Installed it and started to play with it, imported some documentation into memory so my opencode is now better aware of what is it suppose to do.

But I still struggle to understand how to use it properly. Readme only mentioned what is that is behind, configuration, but misses a big point: how to use it on a daily basis?
Do I need to manually ask opencode to remember stuff? What should I store in memory and what in AGENTS.md/documentation files?
How to use "intelligent prompt-based memory extraction"? Good practices?

[rastarr]

@simonachmueller I'm a bit in the same boat as you. nothing appears to be automatic from what I see. no idea what the profile tab is about, what it's used for or even how to populate it. some clear examples in how to actually it opencode-mem insode opencode would be a great starting point since, as you say, it certainly does look like a good plugin

[tickernelz]

Good point — the README focuses on configuration but doesn't have a "how to actually use this day-to-day" guide. This is a valid docs improvement.

A quick answer to your questions:

1. Do I need to manually ask opencode to remember? — No. If autoCaptureEnabled: true (default), the plugin automatically captures relevant information after each conversation turn when the session goes idle. You can also manually use the /memory add <content> command or the memory tool.

2. Memory vs AGENTS.md/documentation — Use memory for: project-specific knowledge, coding decisions, bug patterns, user preferences. Use AGENTS.md for: static instructions that don't change often (project rules, workflow guidelines).

3. Intelligent prompt-based extraction — This is the auto-capture feature. When enabled, after each conversation the AI analyzes the session and extracts memorable information automatically. No manual action needed.

We'll work on improving the documentation. Thanks for the feedback.