Intelligence shouldn't be rented. It should be inherited.
Nemosis is a high-performance framework designed to automate "Knowledge Osmosis" between high-capability Teacher models and cost-effective Student models. Built with Rust and compliant with the AgentSkills.io specification.
Nemosis collapses expert reasoning into a standardized SKILL.md. This file contains the Strict Contract (JSON Schema) and Strategic Context (Markdown Instructions), enabling Student models to inherit complex capabilities without the high cost of multi-turn reasoning.
- Core: Rust (Safety, Performance, Portability)
- Orchestration: pnpm Workspaces (Monorepo management)
- Distribution: npm Wrapper (Seamless CLI integration)
- Standard: AgentSkills.io (
SKILL.md)
- Capture: Pull high-fidelity execution traces from OpenClaw.
- Distill: The Rust-based Distiller analyzes the traces and synthesizes a
SKILL.md. - Osmose: The Student model is onboarded by parsing the
SKILL.md, instantly expanding its toolset. - Audit: The Teacher model evaluates performance and triggers a Rust-driven refactor of the skill if logic drifts.
.
├── pnpm-workspace.yaml
├── package.json # Root scripts
├── packages/
│ ├── nemosis-core/ # Rust: Logic for experience extraction
│ │ ├── src/
│ │ └── Cargo.toml
│ ├── nemosis-cli/ # CLI Entrypoint (npm wrapper)
│ └── nemosis-skills/ # Auto-generated SKILL.md registry
│ └── <skill_id>/
│ └── SKILL.md # The distilled expert skill
└── scripts/ # Build & Wrap scripts
- Rust-Powered: Fast, safe, and compiles to a single binary.
- Spec-First: Built on the
SKILL.mdstandard for maximum interoperability. - Minimalist: No heavy dependencies, just clean, high-efficiency osmosis.