NemoHermes
https://github.com/hmbown/nemohermes- Category
- AI Tools
- Rank
- No. 1699Tools index
Previous survey · No. 1704 ·
- Pricing
- Open Source
- Type
- TOOL
- Builder
- Hmbown
- GitHub
- 8 stars
- Latest release
- v0.1.0
- Date
About
NVIDIA capability registry and Spark-aware routing layer for Hermes Agent and NemoClaw users.
What it does
NemoHermes finds supported AI services running on a local machine, combines them with a built-in NVIDIA catalog, and records their capabilities in a disk cache. Its CLI then selects and explains a suitable service for jobs such as chat, vision, transcription, speech synthesis, training, or hosting.
Why it's ranked here
The project solves a specific operational problem with inspectable decisions rather than hidden endpoint selection. Local discovery, cached state, explicit constraints, and fallback explanations form a coherent workflow. The narrow NVIDIA and DGX Spark focus makes it useful in that environment, but unsuitable as a general backend router.
What's good
Routing considers health, locality, deployment profile, backend preference, modalities, streaming, tool calling, structured output, and realtime support. Operators can inspect discovered services and selection reasons through text or JSON output. A five-minute cache avoids probing on every request and remains available when refresh fails.
Tradeoffs
Discovery only checks fixed local endpoints for five supported runtimes. It does not discover arbitrary hosts or ports. Failed probes still produce registry entries, although routing scores healthy services higher. Built-in catalog entries are marked healthy without a live check, and some represent conceptual workflow slots rather than callable inference endpoints. The project explicitly identifies itself as early alpha.
How to use it well
Use it when Hermes Agent or NemoClaw needs predictable service selection across DGX Spark, NVIDIA workstations, or Mac plus Spark setups. Run discovery, inspect the merged registry, test route decisions with realistic constraints, then connect those decisions to the host workflow. It does not deploy runtimes, execute inference, train models, or provide general-purpose backend integration.
Technical notes+
The TypeScript package exposes a standalone executable through package.json and an OpenClaw-compatible registration bridge in src/index.ts. src/discovery.ts performs HTTP GET probes with 1.5-second timeouts against fixed localhost endpoints for vLLM, SGLang, NIM, faster-whisper, and Piper. src/registry.ts merges probe results with src/catalog.ts, writes JSON state, reuses fresh state, and falls back to existing cached state after refresh errors. src/router.ts filters candidates by hard capability constraints, excludes cloud records when configured, then applies additive scoring for health, reachability, locality, backend preference, realtime support, profile fit, transport, and metadata priority. Tests in src/index.test.ts, src/router.test.ts, and src/registry.test.ts cover configuration normalization, routing behavior, catalog merging, and cache freshness.
Observed
- License
- Apache-2.0
- Primary language
- TypeScript
- Packaging
- npm package with compiled JavaScript and TypeScript declarations
- Install surface
- Runnable through npx; requires Node.js 20 or newer
- Interfaces
- Standalone CLI and OpenClaw-compatible CLI extension
- Local probes
- vLLM, SGLang, NIM, faster-whisper, and Piper
- Testing
- Vitest tests cover configuration, routing, catalog merging, and registry caching
Read from README.md, package.json, src/cli.ts, src/index.ts, src/types.ts, src/router.ts, src/catalog.ts, src/registry.ts, src/discovery.ts, src/index.test.ts, src/standalone.ts, src/router.test.ts, src/registry.test.ts, src/commands/route.ts, src/commands/models.ts.
What it can do
Register NVIDIA capabilities
NVIDIA hardware/software capability specifications → Registered capability entries in registry
Route Spark jobs based on available capabilities
Spark job requirements and available NVIDIA resources → Optimized job routing decisions
Discover available NVIDIA resources
Resource query parameters → List of available NVIDIA capabilities and resources
Manage Hermes Agent connections
Hermes Agent registration requests → Established agent connections and routing rules
Handle NemoClaw user requests
NemoClaw user job submissions → Routed requests to appropriate NVIDIA resources
Monitor resource availability
Real-time system status data → Updated resource availability information
Tags
Tech Stack
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.