Installation
Installing visvoai-ai and visvoai-core, and picking your provider extras
The two packages are independent. Install only what you need.
visvoai-ai
Base install is light — the provider SDKs are extras, so you only pull the dependency tree of the provider you actually call:
pip install "visvoai-ai[gemini]" # langchain-google-genai + google-genai
pip install "visvoai-ai[anthropic]" # anthropic + langchain-anthropic
pip install "visvoai-ai[openai]" # openai + langchain-openai (also covers every
# OpenAI-compatible endpoint: Together, Groq,
# OpenRouter, vLLM, LM Studio, llama.cpp servers)
pip install "visvoai-ai[all]" # every extra aboveAn OpenAI-compatible provider (anything reached via OpenAICompatProvider)
needs no dedicated extra beyond [openai] — it rides langchain-openai
with a different base_url and API key. See
Providers & the model registry.
visvoai-core
pip install visvoai-coreThis pulls only langgraph and langchain-core — no provider SDKs, no
datastore client, no web framework. visvoai-core takes any LangChain
BaseChatModel you already have; pairing it with visvoai-ai is a choice,
not a requirement.
Environment variables
visvoai-ai resolves API keys from the environment by default (override
with api_key= at call time):
| Provider | Environment variable |
|---|---|
gemini | GEMINI_API_KEY |
anthropic | ANTHROPIC_API_KEY |
openai | OPENAI_API_KEY |
together | TOGETHER_API_KEY |
groq | GROQ_API_KEY |
openrouter | OPENROUTER_API_KEY |
Any other OpenAI-compatible provider (vLLM, LM Studio, a models.dev-sourced
provider, …) needs an explicit api_key= and base_url=, or a key_env
carried by a catalog-sourced deployment — see
Providers & the model registry.
visvoai-core needs no environment variables of its own — it has no I/O
boundary beyond the model and tools you give it.
Verifying the install
Listing the model registry works with no API key — it's pure in-package data:
from visvoai.ai import list_deployments, Capability
for d in list_deployments(Capability.CHAT)[:5]:
print(d.id, d.display_name, d.input_cost_per_million)Building and invoking a model needs the matching key set. See
Cost, usage & thinking levels for the registry in
depth, or jump straight to the agent loop for
visvoai-core.