The difference is that this does not build a new memory layer.

OpenCode already stores the full conversation history from every session your agent has ever run across every project on the machine: messages, tool calls, tool outputs, reasoning traces. Even after compaction shrinks what the model can see, the original content stays in the database.

This plugin exposes that hidden history back to the agent.

So instead of creating a separate memory system with embeddings, summaries, extraction, or duplicated storage, it gives the agent direct access to context OpenCode already has but the LLM cannot currently see.

In practice that means:

• recovering tool output and errors that were pushed out of the context window
• pulling back exact original user requirements after long sessions
• searching across older sessions in the same project
• searching across sessions from other projects on the same machine

Memory plugins and this plugin can complement each other, but they solve different problems:

• memory plugins store new curated memory
• opencode-session-recall unlocks dormant history that is already there