Vectors
Embedded similarity search that runs inside your process: persistent collections, metadata filtering applied during the search, ANN indexes, and adapters for Spring AI and LangChain4j. Takes full advantage of the Java 25 Vector API.
Apache 2.0 · MIT
Integrallis OSS
OSS Projects, released under permissive licences and useful on their own. Four compose into the JVM AI runtime; the rest came out of the research and ship on their own terms.
01Projects
Libraries, tools, and a corpus.
Each one exists because a piece of the research needed it. The four at the top compose into a locally owned AI runtime and are documented in depth on the Java AI page; the rest stand alone.
Models
In-process small-language-model inference — no Python, no model server, no sidecar. Pure Java, native FFM, and Apple platform backends, selected from recorded capability evidence. Rides on Vectors.
ModelJARs
Qualified, versioned model artifacts resolved through ordinary JVM dependency coordinates, with provenance and checksums verified against upstream weights. Declare a model the way you declare any other dependency.
Memory
An open agentic memory layer: working and long-term stores, retrieval orchestration, consolidation, and conflict resolution. Framework-neutral at the core, with Spring AI, LangChain4j, and MCP surfaces.
Repository private
Model router corpus
The reference corpus and qualification criteria behind intelligent routing: how candidate models are scored, compared, and published so a router can select on measured capability instead of reputation.
MFCQI
A normalized, evidence-based code-quality score designed to act as a cheap canary in fast or autonomous development loops — keeping generated code from quietly compounding technical debt.
LangGOAP
Goal-oriented action planning for agents using deterministic A* search, with optional constraint optimization and automatic replanning when a step fails mid-execution.
LLMTivo
Record and replay model interactions so an evaluation, a regression test, or a demo is hermetic and repeatable rather than re-billed and re-rolled on every run.
pleat
Ribbon filters with pleated construction: a one-pass, cache-window reordering that builds space-optimal filters at close to Bloom-filter speed, roughly twice as fast as building in arrival order at scale. Companion to the Ribbon Catches Bloom paper.
Build locally