AI & LLMs
Three distinct layers get lumped together here: the model providers themselves, the SDKs that give you one interface across them, and the runtimes for running open models on your own hardware.
7tools
- AI SDKOne TypeScript interface for streaming, tools and structured outputai-sdk.dev
- Claude APIAnthropic's models, with strong long-context and tool-use behaviouranthropic.com
- OllamaRun open models on your own machine with one commandollama.com
- OpenAI PlatformThe most widely integrated model API, and the de facto request formatplatform.openai.com
- OpenRouterOne API and one bill across hundreds of models from every provideropenrouter.ai
- ReplicateRun open models behind an API without provisioning a single GPUreplicate.com
- StashbaseScoped secrets and API access for coding agents, without handing them the keysstashbase.dev
Head to head
AI & LLMs, compared
Read the Layer column first — a provider, a router, an SDK and a local runtime are complements, not rivals.
| Tool | Layer | Inference runs | Billing | Locks you to |
|---|---|---|---|---|
| AI SDK | SDK | Whichever you point it at | Free | Nothing |
| Claude API | Model provider | Hosted | Per token | One provider |
| Ollama | Local runtime | Your machine | Free | Nothing |
| OpenAI Platform | Model provider | Hosted | Per token | One provider |
| OpenRouter | Router | Hosted | Per token, plus margin | One gateway |
| Replicate | Model host | Hosted | Per second of compute | One host |
| Stashbase | Agent access layer | Not involved | Per seat | Its secrets store |