# I need to use LLM inference APIs

> Hosted and local inference APIs, most of them OpenAI-compatible, that serve open, proprietary and on-device models.

The best fits to use LLM inference APIs on Agent Switchboard, as of 20 Nov 2026: NVIDIA NIM API, Crusoe Managed Inference and Prime Inference, of 54 tools that do it.

## Quick answers

- **Which official tools use LLM inference APIs?** NVIDIA NIM API, Crusoe Managed Inference and SambaNova Cloud (SambaCloud), among 35 that are official.
- **Which have a hosted MCP server?** None of the 54 tools lists a hosted MCP server address.
- **Which need no API key or sign-in?** Magnitude, Jan and vLLM, among 13 that need none.

## Best fits

1. [NVIDIA NIM API](https://agentswitchboard.dev/agents/nvidia-nim-api) (api · Official): OpenAI-Compatible Chat: Runs chat completions on hosted open models at integrate.api.nvidia.com/v1, with SSE streaming and tool calling.
2. [Crusoe Managed Inference](https://agentswitchboard.dev/agents/crusoe-managed-inference) (api · Official): Run open-source and fine-tuned models through Crusoe Cloud's OpenAI-compatible inference API, on serverless per-token endpoints or dedicated self-serve deployments.
3. [Prime Inference](https://agentswitchboard.dev/agents/prime-inference) (api, cli): Serves hosted open models and gateway access to third-party models through one OpenAI-compatible API, with rerank and classify endpoints and a CLI.
4. [SambaNova Cloud (SambaCloud)](https://agentswitchboard.dev/agents/sambanova-cloud) (api · Official): Serves open models fast on SambaNova chips through an OpenAI-compatible inference API and Python SDK.
5. [Cerebras Inference](https://agentswitchboard.dev/agents/cerebras-inference) (api · Official): Serves open-weight LLMs such as Qwen and GPT OSS through an OpenAI-compatible API on wafer-scale hardware, for developers who need very fast inference.
6. [Nebius Token Factory](https://agentswitchboard.dev/agents/nebius-token-factory) (api · Official): Runs inference and fine-tuning on open models through Nebius's OpenAI-compatible API, with sandboxes for agents.
7. [Kimi API Platform](https://agentswitchboard.dev/agents/kimi-api) (api · Official): Serves Moonshot AI's Kimi models through OpenAI-compatible Chat Completions, Responses, and Anthropic-compatible Messages APIs, with tool calling, web search, and a 1M-token context.
8. [Magnitude](https://agentswitchboard.dev/agents/magnitude) (cli, api): Local Inference API: Serves OpenAI Chat Completions, OpenAI Responses, and Anthropic Messages routes on port 10100, as JSON or server-sent event…
9. [OpenAI API](https://agentswitchboard.dev/agents/openai-api) (api, cli · Official): Text Generation: Generates text, code, and structured JSON from prompts with OpenAI models through the Responses API.
10. [Groq](https://agentswitchboard.dev/agents/groq) (api · Official): Low-latency inference API running open models on custom LPU hardware, with an OpenAI-compatible interface and tool-calling support.
11. [Fireworks AI](https://agentswitchboard.dev/agents/fireworks-ai) (api · Official): Production inference platform for open models with fast function calling, structured output, and fine-tuning behind an OpenAI-compatible API.
12. [Jan](https://agentswitchboard.dev/agents/jan) (api, cli): Runs local LLMs 100% offline in a desktop app and connects cloud models, exposing a local OpenAI-compatible API. Privacy-first ChatGPT alternative by Menlo Research.

All 54, with filters: https://agentswitchboard.dev/browse?task=call-llm-apis

## For agents

- This task as JSON: https://agentswitchboard.dev/api/tasks/call-llm-apis
- One pick for your user's need, with the next step: https://agentswitchboard.dev/api/recommend?need=use%20LLM%20inference%20APIs (or the `recommend` MCP tool at https://agentswitchboard.dev/api/mcp)
- Every task: https://agentswitchboard.dev/tasks · the whole site: https://agentswitchboard.dev/llms.txt

Updated 20 Nov 2026 with the weekly drop. Ranked by how well each listing matches the task, never by payment. Page: https://agentswitchboard.dev/tasks/call-llm-apis
