n8n-nodes-charm-hyper

An n8n community node package for Charm Hyper, the fast, cost-effective inference API for agentic coding.

Hyper speaks both OpenAI and Anthropic wire formats, so this package exposes all three inference surfaces plus account endpoints behind a single credential, and ships a chat model sub-node for use with n8n's AI Agent.

Install

In n8n (GUI)

1. Go to Settings → Community Nodes → Install.
2. Enter n8n-nodes-charm-hyper and confirm.

Manually

cd ~/.n8n
npm install n8n-nodes-charm-hyper

Then restart n8n.

Credentials

Create a Charm Hyper API credential:

| Field | Description | | ------------ | ------------------------------------------------------------------------------------------------------------------------- | | API Key | A key from the Hyper dashboard, starting with sk-hyper-. Sent as Authorization: Bearer …. | | Base URL | Defaults to https://hyper.charm.land/v1. Only change this for a proxy or gateway. |

Credential testing calls GET /v1/credits, so a saved credential with a green check is a working key.

Operations

| Resource | Operation | Endpoint | | ------------------- | --------- | ---------------------------------------- | | Chat Completion | Complete | POST /v1/chat/completions (OpenAI) | | Response (OpenAI) | Create | POST /v1/responses (OpenAI Responses) | | Message (Anthropic) | Create | POST /v1/messages (Anthropic Messages) | | Model | List | GET /v1/models (public) | | Credit Balance | Get | GET /v1/credits |

Chat Completion

The standard OpenAI chat endpoint. Add a Model, build a Messages list, and optionally tune behaviour under Options (temperature, penalties, response format, stop sequences, max output tokens).

Extra OpenAI parameters that are not exposed in the UI can be passed through Additional Body Parameters as JSON. That object is merged into the request body last, so it can also override any field above:

{ "seed": 7, "logit_bias": { "50256": -100 } }

Response (OpenAI)

The newer Responses API. Choose Text for a single prompt or Messages for a full conversation, and optionally supply Instructions as a system-level directive.

Message (Anthropic)

The Anthropic Messages format. max_tokens is required by that API and is exposed as Max Tokens. Use System for the system prompt and Top K / Top P for sampling.

Note: reasoning models may return a thinking content block before the text block. Increase Max Tokens if a response comes back with stop_reason: max_tokens and no visible text.

Model / Credit Balance

Model: List returns the full catalogue including per-model context windows and pricing. It is a public endpoint and needs no authentication. Credit Balance: Get returns your remaining Hypercredits.

Output

Each operation returns the raw API response on the item's json, unmodified. For chat completions that includes Hyper's cost extensions:

{
	"id": "chatcmpl-…",
	"choices": [{ "message": { "role": "assistant", "content": "pong" }, "finish_reason": "stop" }],
	"usage": {
		"prompt_tokens": 38,
		"completion_tokens": 18,
		"total_tokens": 56,
		"cost": { "usd": 0.0001, "hypercredits": 0.002 },
		"remaining": { "hypercredits": 119.5788448 }
	}
}

Read a reply with an expression such as:

{{ $json.choices[0].message.content }}

Using it as a language model

The package also ships Charm Hyper Chat Model, a language model sub-node. Attach one to the Model input of an AI Agent (or any chain that accepts a language model) to run that agent on Hyper. Pick a Model; sampling and request options live under Options.

Because Hyper speaks the OpenAI wire format, the sub-node delegates to n8n's own OpenAI client. Streaming, tool calling and structured output therefore behave exactly as they do with the built-in OpenAI model. Two things worth knowing:

• Use Responses API switches calls to /v1/responses. Some reasoning models require it. While it is on, pass max_completion_tokens through Additional Body Parameters instead of using Max Output Tokens, which maps to max_tokens on the chat completions API.
• The Model dropdown is populated from the public /v1/models endpoint, so the credential is only exercised when the model actually runs. You can also type a model ID with an expression.

Using it as a tool

The main node sets usableAsTool, so it can be attached directly to an n8n AI Agent as a callable tool.

Development

npm install
npm run build     # compile TypeScript to dist/
npm run dev       # live-reload into a local n8n
npm run lint      # lint against the n8n community node rules
npm run lint:fix

Source layout:

credentials/CharmHyperApi.credentials.ts        # API key + base URL, Bearer auth, /credits test
nodes/CharmHyper/
  CharmHyper.node.ts                            # node definition + execute loop
  descriptions.ts                               # resource/operation parameter definitions
  actions.ts                                    # per-operation request building
  CharmHyper.node.json                          # n8n codex metadata
nodes/CharmHyperChatModel/
  CharmHyperChatModel.node.ts                   # language model sub-node (supplyData)
  CharmHyperChatModel.node.json                 # n8n codex metadata
nodes/shared/GenericFunctions.ts                # HTTP transport, model loadOptions, helpers
icons/                                          # light + dark node icons

Links

• Hyper documentation
• Hyper API authentication
• n8n community nodes documentation

License

MIT