Wolbarg

About Wolbarg

Wolbarg is durable shared memory for AI agents — remember something once, then recall it by meaning across sessions, tools, and multi-agent workflows. The problem it targets is familiar to anyone running several coding agents: Cursor, Claude Code, and Codex each drop context when a session ends, so every session re-explains the same architecture, and a decision made in one tool never reaches the others. Multi-agent setups improvise shared state out of chat transcripts, globals, or Redis, and lose semantic search, concurrency safety, and crash safety in the process. Wolbarg ships two things. Wolbarg Workspace is the product for coding agents — install once per repository with `npx @wolbarg/workspace init`, connect the hosts you use via `connect cursor`, `connect claude`, or `connect codex`, and every agent reads the same reconciled project briefing and writes back what it learns. Crucially, updates supersede stale facts rather than piling up as append-only notes, and decisions and discoveries stay tied to the files, tests, and docs that evidence them. A single CLI covers init, connect, doctor, status, and review, with no hand-edited MCP JSON. Underneath is an SDK: an agent calls remember() with a fact and optional metadata, the text is embedded through your own OpenAI-compatible endpoint, the original text, metadata, and vector land in a shared store, and any sibling agent can query in natural language and get ranked hits. Storage is local-first SQLite by default with Postgres when you need it, and AI SDK adapters cover custom agent frameworks on the same memory layer.

Key Features

Wolbarg Pros & Cons

Who Is Wolbarg Best For?

Tags