FAQs

How is Brain different from a vector database or a RAG pipeline?

A vector database is a retrieval primitive: you build and operate the pipeline around it — chunking, embedding, prompts, permissions, and every client integration. Brain is the finished system on top: knowledge is grouped by entity and distilled into compact summaries that cite their source documents and rebuild when a source changes. Agents don't get raw chunks back — they get curated, cited knowledge sized for a context window. And it isn't only retrieval: the same server carries versioned skills and agent-to-agent handoffs, over the one protocol every client already speaks.

How is it different from a model vendor's built-in memory?

Vendor memory lives inside one product and stays there. Brain is one memory shared across Claude, Codex, Obsidian and any MCP client — switch models or providers and the memory comes with you. Knowledge exports out; nothing is trapped in a vendor's memory feature.

Which clients are supported?

Claude Code and Codex install from the GitHub marketplace (levirge/brain); Claude Desktop installs from Settings → Plugins (step-by-step); Obsidian gets search, publish and two-way sync via the levirge/brain-obsidian plugin; and any other MCP client connects to https://brain.levirge.com/mcp over Streamable HTTP (definition). Humans use the web UI.

How do agents authenticate?

OAuth, negotiated when the client connects. Per-user bearer tokens are the fallback — you issue and revoke them yourself, and Brain stores only the token's hash, never the raw credential.

Where does my knowledge go?

Your workspace is tenant-isolated, whichever model lane is active. In private mode, every model call — categorisation, summarisation, chat — runs on LLMs Levirge hosts on its own infrastructure; no cloud model provider is in the path. Hosted mode uses cloud LLMs when speed matters more than locality. Embeddings and reranking run on Levirge infrastructure in both modes.

Can I self-host it?

Not today — we host it. Self-hosting is a roadmap conversation for enterprise; if Brain needs to run somewhere specific, talk to us.

What does it cost?

Brain is in early access. Pricing is quoted per workspace and set up on the access call — request access.

What happens if the same thing is captured twice?

Identical content deduplicates on a content hash, so re-running an ingest or retrying a failed capture is safe.

What happens when a source document changes?

The summaries built from it rebuild, and their citations stay pointed at the documents each statement came from — the distillation doesn't drift from the source.

What if an agent is offline when a handoff arrives?

Nothing is lost: handoffs that arrive while a watcher is down are delivered on its next watch. The inbox watcher wakes a connected agent the moment a handoff lands, with the sender's context attached.

What is a Vault?

The unit of partitioning: knowledge belongs to a Vault, Vaults belong to your Tenant, and access is per-Vault (viewer or member) plus tenant-wide admin. Personal Vaults are private to their owner and fail closed.