nanograph 1.1.2

Multimodal is here. nanograph now handles images, audio, video, and PDFs as first-class graph citizens — external media URIs with vector embeddings you can search and traverse like any other node.

Under the hood, the runtime and query engine have been substantially reworked. Reads are lazy and metadata-first. Relational query tails run through DataFusion. Storage has been refactored with v4-ready seams, and derived Lance mirror tables now materialize committed graph history. Both SDKs ship with full media support.

This is the canonical stable release for the current Lance v3 line — it supersedes v1.1.0 and v1.1.1.

What shipped

Multimodal media nodesImages, audio, video, and PDFs as first-class graph nodes. External URIs, text-to-image retrieval, and traversal from matched media.Blobs guide →

Gemini embeddingsMultimodal embeddings via gemini-embedding-2-preview. Text and media sources, document/query role separation, auto-detection.Embeddings guide →

Faster runtimeLazy metadata-first opens, DataFusion-backed queries, and a cleaner write path. Less memory, faster reads.

SDK media supportTypeScript and Swift SDKs ship with typed media ingest, embed(), and text-to-image retrieval.

v4-ready storageWAL and graph history now materialize into derived Lance tables. Storage split behind neutral version interfaces, preparing for the Lance 4 transition.

Upgrading

brew upgrade nanograph or grab the binary from the release. Rust crates are published on crates.io for nanograph, nanograph-cli, nanograph-ffi, and nanograph-ts.

New databases default to Lance v3 / storage format 2.2. Existing 2.0 databases remain readable. See the migration guide if you want to upgrade storage format.

Skills updated

The nanograph-skills package has been updated for 1.1.2. Both skills now cover the new features:

nanograph-opsNow covers blobs and media nodes, search predicates (lexical, semantic, hybrid with bm25/rrf), embedding providers (OpenAI + Gemini), and new output formats.Skills docs →

nanograph-lanceUpdated with compression benefits of format 2.2, clearer Lance v3 vs format distinction, and improved migration steps.

New docs

Blobs and Media Nodes — media storage, @media_uri, JSONL formats, and media-root behavior
Embeddings — providers, @embed, text and multimodal vectors, hybrid search
Skills — install agent skills so your AI tools know how to operate nanograph

Star the repo if you want to follow along.