Your coding agent finally gets a memory that outlasts the session

What it does

opencode-mem is an OpenCode plugin that stores facts, decisions, and context from your AI coding sessions in a local vector database built on SQLite. It retrieves relevant memories automatically when you prompt the agent, and scopes everything either to the current project or across all projects. A built-in web UI runs locally for visual timeline browsing and memory management.

The interesting bit

Instead of requiring its own API keys, the plugin piggybacks on OpenCode’s existing provider authentication to extract memories from prompts—routing through whatever you’ve already set up, whether that’s Claude via OAuth or a direct OpenAI key. It also exposes canonical tagging helpers (opencode-mem/tags) so other plugins can write into the same memory store without accidentally creating orphaned shadow shards.

Key highlights

• Local-first: USearch vector indexing with SQLite as the source of truth, falling back to exact scan if USearch fails
• Auto-capture feeds prompts through your existing OpenCode AI provider to extract structured memories—no extra API key needed
• 12+ local embedding models available, plus smart deduplication and automatic user profile learning
• Built-in web UI for browsing the memory timeline and managing what the agent remembers
• Cross-plugin memory store with stable subpath exports for third-party tools

Verdict

Worth a look if you use OpenCode daily and are tired of re-explaining your project structure to an agent every session. If you are not already inside the OpenCode ecosystem, this plugin has nothing to offer you.