Cloud Providers

You only need an API key for the providers you actually use — SynapseOrch picks the provider for each call from the model name.

Quick reference

Settings → Models groups the cards in three: Model makers (OpenAI, Anthropic, Gemini, Grok, DeepSeek, Mistral, Moonshot, Qwen, Z.ai, MiniMax), Hosts (Bedrock, OpenRouter, Groq, Cerebras, Together AI, Fireworks, NVIDIA) and On your machine or server.

More providers, each with its own key

Eleven more providers have a card of their own. Each takes an API key, lists its models from that key, and calls tools natively. A model is named with the provider's prefix, which SynapseOrch strips before the call:

Groq and Cerebras answer fastest, which suits a phone agent.

On SynapseOrch Cloud the same cards are under Settings → Models → Providers, grouped as Model makers, Hosts and Your own server. A model on your own key is billed by your provider, not from the model wallet.

For local providers (Ollama, vLLM, LM Studio), see Local Providers. SynapseOrch's own hosted models are the model gateway.

Saving. Each card saves its key with the Save beside the key field; its other fields — a region, a base URL, model names — save with the Save at the foot of the card. A card reads No key configured until it has a key, then the number of models it lists. If a provider rejects the key, its model list falls back to a built-in list, and calls fail with the provider's error.

Anthropic (Claude)

Model names: claude-…

1. Open Settings → Models and click the Anthropic Claude card.
2. Paste your API key into API key (format: sk-ant-...). The card links to the Anthropic Console if you need to generate one.
3. Click Save. Claude models appear in the model selectors — the list is whatever models your key can see.

For example:

The key is stored as the anthropic_key setting.

OpenAI (GPT)

Model names: gpt-…, chatgpt-…, and the o-series — o1, o3, o3-mini, o4-mini

In Settings → Models click the OpenAI card, paste your key (format sk-...) into API key, and click Save.

Model names beginning with gpt or chatgpt, or with o and a digit, are sent to OpenAI. The selector lists every chat model your key can use — gpt-5…, gpt-4.1, gpt-4o, o1, o3, o3-mini, o4-mini, chatgpt-4o-latest and the rest. It leaves out what the chat API cannot serve: audio, image, realtime, text-to-speech, transcription, moderation, search and embedding models, and the -pro, codex and deep-research models, which only OpenAI's Responses API serves. The key is stored as the openai_key setting.

Google Gemini

Model names: gemini-…, gemma-…

In Settings → Models click the Google Gemini card, paste your key (format AIza...) into API key, and click Save. The card links to Google AI Studio, where you can get a free key. The selector lists every Gemini model your key can call.

The key is stored as the gemini_key setting.

xAI (Grok)

Model names: grok-…

In Settings → Models click the xAI Grok card, paste your key (format xai-...) into API key, and click Save.

The key is stored as the grok_key setting.

DeepSeek

Model names: deepseek-…

In Settings → Models click the DeepSeek card, paste your key (format sk-...) into API key, and click Save.

DeepSeek models don't accept images. The key is stored as the deepseek_key setting.

AWS Bedrock

Model names: bedrock.…

Bedrock exposes Anthropic, Meta, Mistral, and Amazon foundation models through one AWS endpoint. In Settings → Models click the AWS Bedrock card and fill:

Models are referenced with the full Bedrock model ID prefixed by bedrock.:

bedrock.anthropic.claude-sonnet-4-20250514-v1:0
bedrock.meta.llama3-8b-instruct-v1:0
bedrock.mistral.mistral-large-2402-v1:0

Inference profiles are listed as models — the us., eu., apac. and global. variants and your application profiles. Pick one wherever you choose a model; a model with no on-demand throughput is only available that way.

Access keys also work instead of a Bedrock API key: set the aws_access_key_id and aws_secret_access_key settings (and aws_session_token for temporary credentials). They have no field on the card.

Embedding models on Bedrock

Bedrock's embedding models, such as bedrock.amazon.titan-embed-text-v2:0, appear in the separate Embedding model selector under Defaults in Settings → Models — the Bedrock card has no embedding field. The embedding model indexes code repositories and matches each step's semantic cache. Keep the bedrock. prefix.

The card's fields are stored as bedrock_api_key and aws_region.

OpenAI-Compatible (Cloud)

For cloud providers that expose an OpenAI-compatible API — OpenRouter, Together AI, Fireworks AI, Anyscale, etc.

Model names: oaic.<model_name>

In Settings → Models click the OpenAI Compatible card and fill:

The card is enabled once it has a Base URL. If the provider answers /v1/models, SynapseOrch lists those models and Model names is not needed. Use a model in agent settings as oaic.meta-llama/llama-3-70b-instruct (prefix + the model's name).

The fields are stored as openai_compatible_key, openai_compatible_base_url, openai_compatible_models and openai_compatible_embed_models.

Other model settings on this screen

• Default model — used by every agent and step that doesn't pick its own. See Model Providers.
• Builder model — the model Synapse AI uses.
• Embedding model — indexes code repositories, matches each workflow step's semantic cache, and embeds knowledge bases unless Knowledge has its own.

Prompt caching

All cloud providers above benefit from the prompt cache, which is on by default. On a self-hosted install, Usage's Cache analytics shows the savings, without any per-provider configuration.