A portable LLM kernel. One surface, many providers, intent-driven routing.
xLLM is a thin TypeScript adapter layer over LLM providers. It exposes an
OpenAI-compatible wire format and routes requests by intent —
sovereign, powerful, cheap, fast, or balanced — to the concrete
provider/model that best matches. Swap OpenAI for Anthropic for a local
Ollama server without changing application code.
v0 — spec stage. See specs/v0.md for the design.
The landing page lives in apps/www.
| Intent | Priority | Example backend |
|---|---|---|
sovereign |
Runs locally, private | SovereignLocalAI, Ollama |
powerful |
Highest quality | Claude Opus, GPT-5 |
cheap |
Lowest cost | Haiku, 4o-mini (via OpenRouter) |
fast |
Lowest latency | Groq, Cerebras |
balanced |
Quality per dollar | Sonnet 4.6, GPT-4.1 |
- OpenAI
- Anthropic
- OpenRouter (cloud multiplexer)
- Ollama (local)
- SovereignLocalAI (local)
- Custom (any OpenAI-compatible endpoint)
xLLM.WTF/
├── apps/
│ └── www/ # Next.js landing page
├── packages/
│ └── core/ # @xllm/core — provider interface + adapters (planned)
└── specs/
└── v0.md # v0 design spec
Apache 2.0