opencode-mem

Provides coding agents with a local vector database backend that persists agent interactions, code context, and learned patterns across sessions without requiring external cloud infrastructure. Uses embeddings to store and retrieve contextual information, enabling agents to maintain continuity and reference past decisions without re-processing the same codebase analysis.

Retrieves semantically similar code snippets and architectural patterns from the agent's memory using vector similarity search, allowing agents to find relevant past solutions without keyword matching. Converts code and documentation into embeddings, then performs nearest-neighbor queries to surface contextually relevant information for code generation tasks.

Automatically captures and stores agent decisions, code generation choices, and reasoning steps in the vector database, creating a queryable history of what the agent has done and why. Each decision is embedded and indexed, allowing agents to review their own past reasoning patterns and avoid repeating failed approaches.

Manages a local vector database instance that stores embeddings, metadata, and retrieval indices without external dependencies. Handles database initialization, embedding storage, index management, and query execution entirely on the developer's machine, with built-in support for persistence across restarts.

Integrates seamlessly with the OpenCode framework as a plugin, exposing memory and retrieval capabilities through OpenCode's standard plugin API. Handles lifecycle management, configuration, and inter-plugin communication, allowing coding agents built on OpenCode to access memory features without custom integration code.

Converts code snippets into vector embeddings and performs similarity matching to find structurally and semantically similar code patterns. Uses embedding models to capture code semantics beyond syntax, enabling agents to identify related code even when written differently, and rank results by relevance score.