What base URL does Dify need for NovAI?

https://aiapi-pro.com/v1 - add it in Dify under Settings > Model Providers > OpenAI-API-compatible as the API base URL together with your NovAI key; the model list is fetched from /v1/models.

Are there free models I can use in Dify workflows?

Yes: GLM-4.7-Flash, GLM-4.6V-Flash, GLM-4.1V-Thinking-Flash, CogView3-Flash, CogVideoX-Flash are free and unlimited on NovAI (no token cap, no card). New accounts also get $2 credit for paid models, so you can prototype a whole workflow for free.

Does NovAI add a markup on model prices?

No. Prices are the raw provider list rate with 0% platform fee (vs OpenRouter ~5.5%). Live per-model prices: /china-ai-leaderboard.json.

Dify + Chinese AI Models (2026)

Updated 2026-10-10 · verified against the live model list

Dify is the leading open-source LLM workflow platform — visual canvas, RAG pipelines, agent nodes and multi-step orchestration. Because it accepts any OpenAI-compatible provider, you can plug NovAI into Dify and build workflows on 121 Chinese frontier models — DeepSeek, Qwen, GLM, Kimi, Seedance, MiniMax, Hunyuan — with five of them free and unlimited.

Connect NovAI to Dify

In Dify open Settings → Model Providers.
Choose OpenAI-API-compatible and click Add Model.
Set API Base URL to https://aiapi-pro.com/v1 and API Key to your NovAI key.
Save. Dify fetches /v1/models and every Chinese model appears in the model selector for your workflow nodes.

# optional: self-host Dify via docker-compose
git clone https://github.com/langgenius/dify.git
cd dify/docker && cp .env.example .env
docker compose up -d
# then add the NovAI provider in the web UI at http://localhost

Which model per node

Workflow node	Model	Cost
Classifier / router (high volume)	glm-4.7-flash	Free, unlimited
Vision / image understanding	glm-4.6v-flash	Free, unlimited
RAG answer generation	glm-5.3-flash	$0.08 / $0.28 per 1M
Agent reasoning / tool calls	deepseek-v4-pro / kimi-k2.7-code	see live leaderboard

Workflow patterns that save money

Router → specialist: use a free flash model to classify intent, then escalate only hard queries to a paid pro model. Most tokens stay on the free tier.
Draft → refine: generate a first draft with a free model, then a single refinement pass with a pro model.
Batch summarisation: run high-volume summarisation on a free unlimited model; escalate only edge cases.

Troubleshooting

Model list empty: make sure the base URL ends with /v1 and the key has no surrounding spaces; test curl https://aiapi-pro.com/v1/models.
401: key not active yet or credits exhausted — top-up from $5, or switch to a free model.
Vision node fails: confirm the model is a vision model (glm-4.6v-flash is free and unlimited) and that the node passes an image input.

Exact live prices for all 121 models: leaderboard (machine-readable JSON). NovAI charges 0% platform fee — you pay the raw provider list price.

Get a key: $2 free credit · Browse models: model catalog · Pricing: live leaderboard