Run models locally
Find MLX models that fit your Apple Silicon host, verify them against an installed runtime, and wire a selected model into an agent workflow.
MLX Agent + mlx-scoutExplore local modelsindependent research / open source
Open tools for agents that work across runtimes, editors, and environments.
CAVI builds inspectable tools for local models, agent development, and knowledge workflows.
/featured-outcomes
Start with what you want to accomplish
Three short paths into the tools, packages, and workflows behind the work.
Build across agent runtimes
Use typed packages and explicit boundaries to connect compatible agent runtimes.
API client + runtime providersBrowse packagesWork with knowledge in Obsidian
Pair an in-vault companion with focused agent skills for research and writing.
Obsidian plugins + focused skillsExplore plugins/catalog/latest
Featured releases
A compact starting set drawn from checked-in release manifests, editorial placement, and versioned documentation.
- pluginMLX Agent v0.3.0Discover, adopt, and safely wire local MLX models on Apple Silicon.
- package@cavi-ai/api-clientTyped integration surfaces with versioned technical documentation.
- skillmlx-scout"Discover MLX-optimized models on HuggingFace suited to this Apple Silicon host, routed by role (general, coding, reasoning, vision, embedding)."
/research
Research principles
Research starts in working software with assumptions and evidence that anyone can inspect.
contracts.verify()
Inspect the boundary
Prefer explicit contracts and executable checks over implied behavior.
systems.interoperate()
Cross runtimes carefully
Make tools portable without pretending upstream contracts belong to us.
workflows.stay_local()
Keep local control
Meet people where their models and knowledge already live.