Run Flowise flows on any catalog model via one Base Path.

Quick answer: credential, Base Path, model id.

In the Flowise canvas, drag a ChatOpenAI node from the Chat Models panel, click Connect Credential, and create a new OpenAI credential holding your APIsRouter key. Then open Additional Parameters on the node and set Base Path to https://api.apisrouter.com/v1. Every call the node makes now goes to the gateway instead of OpenAI's hosts. For the model field: the standard ChatOpenAI node offers a dropdown of OpenAI model names, so for gateway ids like claude-sonnet-4-6 or deepseek-v4-flash use the ChatOpenAI Custom node instead, which accepts any model id as free text. This is a documented Flowise pattern, described in the official docs for exactly this class of OpenAI-compatible endpoint.

Connect Credential:  new OpenAI credential
                     API Key: sk-YOUR-APISROUTER-KEY

Model Name:          claude-sonnet-4-6

Additional Parameters:
  Base Path:         https://api.apisrouter.com/v1

How Flowise talks to the endpoint.

Flowise (FlowiseAI on GitHub, roughly 55K stars) is the visual builder for LLM apps: chatflows and agent flows assembled from nodes on a canvas, covering chat models, RAG pipelines over document stores, tool-using agents, and API endpoints you can call from your own products. Every LLM node in a flow issues standard chat-completions requests, and the Base Path decides where those requests land. Because the gateway serves many vendors behind one URL, one credential covers every model in the catalog. A RAG chatflow can run its answer synthesis on claude-sonnet-4-6 while a classification branch in the same flow runs deepseek-v4-flash, each configured as its own node with its own model id, both billing through the same key. The docs describe the same Base Path mechanism for other OpenAI-compatible services, so this is standard Flowise usage rather than a workaround. One honest caveat: embeddings are a separate node type. If your flow uses OpenAI embeddings, configure the OpenAI Embeddings Custom node with the same Base Path pattern, and verify the embedding model id against the catalog first, since chat and embedding surfaces are distinct.

Self-hosted Flowise: Docker networking notes.

Most Flowise installs run in Docker, and the Base Path is resolved from inside the container. A public gateway URL like https://api.apisrouter.com/v1 works from anywhere, so no special networking applies; this caveat exists because much of the community documentation was written for local inference servers, where host.docker.internal addresses matter. Against a hosted gateway you can ignore that entire class of advice. The part that does matter operationally: credentials in Flowise are stored server-side and shared across flows, so one APIsRouter credential created once is reusable by every node in every flow on the instance. For teams, that plus per-key usage in the gateway console gives you a clean split: one credential per environment or per project, and the console attributes tokens and spend to each.

curl -s https://api.apisrouter.com/v1/models \
  -H "Authorization: Bearer $APISROUTER_API_KEY" | head -30

Choosing models per node.

Flows multiply model calls in ways a single chat never does: a RAG pipeline with query rewriting, retrieval grading, and answer synthesis can hit the LLM three or more times per user message. Putting the flagship only where it earns its rate, and a volume tier everywhere else, is a per-node dropdown decision once the Base Path is set.

• claude-sonnet-4-6 for the main conversational or agent node, where answer quality carries the flow.
• claude-haiku-4-5-20251001 for high-volume branches: routing, tagging, short summaries inside larger flows.
• gpt-5.4-mini as the GPT-family volume pick for flows built around OpenAI-style behavior.
• gemini-3.5-flash where a branch ingests long documents or images at a mid-tier rate.
• deepseek-v4-flash for the highest-volume utility calls, where the per-token rate is the design constraint.

Pay-as-you-go · transparent per-model pricing

Selected models are priced below official list prices. Exact input, output, cache, and per-request prices are shown for each model.

Failure modes specific to Flowise.

The two-command debug order applies here as everywhere: prove the endpoint with curl first, models listing and then one completion with the exact id, and only then debug inside Flowise. That separates gateway problems from canvas configuration in under a minute.

• Model dropdown does not list your id: expected on the standard ChatOpenAI node. Use ChatOpenAI Custom, which takes free-text model ids.
• 401 from the node: the credential holds the wrong key, or a different credential is attached to this node than you think. Credentials are per-node attachments, so check the node, not just the credential store.
• 404 on every call: Base Path is missing /v1 or has a trailing path segment it should not have. The node appends route paths to the base you give it.
• Works in one flow, fails in another: each node carries its own Base Path setting under Additional Parameters. Copying a node copies it; building a fresh node does not.
• Embedding nodes still hitting OpenAI: chat and embedding nodes are configured independently. Set the Base Path on the embeddings node too, or expect mixed billing.

Who routes Flowise through a gateway.

• Builders shipping client chatbots who want model choice per flow without a vendor account per client.
• RAG-heavy teams whose retrieval pipelines multiply per-message LLM calls and need volume tiers on the utility steps.
• Agencies running one Flowise instance for many projects, using one credential per project key for clean per-client usage attribution.
• Developers comparing Claude, GPT, and DeepSeek on the same flow, where each candidate is a model-field edit on a duplicated node.
• Self-hosters who want the whole instance billed from one prepaid balance instead of several vendor cards.

FAQ