Naagmani provides native adapters for all leading commercial model providers as well as self-hosted open-source inference runtimes.

Rendering diagram...

graph LR
    GW["Naagmani Gateway"] --> OAI["OpenAI (GPT-4o, o3-mini)"]
    GW --> Anth["Anthropic (Claude 3.5 Sonnet, Haiku)"]
    GW --> Gem["Google Gemini (2.5 Flash, Pro)"]
    GW --> DS["DeepSeek (V3, R1)"]
    GW --> Self["Self-Hosted (vLLM, Ollama, TGI)"]

Supported Providers #

Provider	Supported Models	Protocol Capabilities
OpenAI	`gpt-4o`, `gpt-4o-mini`, `o1`, `o3-mini`, `text-embedding-3-small`	Chat, Streaming SSE, Tool Calling, Vision, JSON Schema
Anthropic	`claude-3-5-sonnet-20241022`, `claude-3-5-haiku-20241022`, `claude-3-opus`	Chat, Streaming SSE, Tool Calling, Vision
Google Gemini	`gemini-2.5-flash`, `gemini-2.5-pro`, `gemini-1.5-flash`	Chat, Streaming SSE, Tool Calling, Multimodal
DeepSeek	`deepseek-chat` (V3), `deepseek-reasoner` (R1)	Chat, Reasoning Streams, Low Cost Inference
Self-Hosted	Any HuggingFace model hosted via vLLM, Ollama, or TGI	OpenAI-compatible endpoint translation

Universal Protocol Normalization #

Every provider utilizes different parameter formats, role definitions, and error schemas. Naagmani's gateway engine normalizes:

Message roles (system, user, assistant, tool).
Function and tool call schemas.
Server-Sent Events (SSE) delta structures.
Upstream HTTP error codes and overload responses.

Next Steps #

Learn about model aliases: Models & Virtual Aliases
Configure provider keys: BYOK & Credential Pools

Model Providers & Adapters

Supported Providers #

Universal Protocol Normalization #

Next Steps #