Vendo Docs

Gemini family of models. Vendo holds the upstream credential and bills per token (including cached input) through the proxy with a route allowlist on /v1beta/models.

Drop-in Gemini API. Point the official Google Generative AI SDK at Vendo's proxy and Vendo authenticates the call, debits credits per token, and forwards to Google. No separate Google Cloud project, no separate billing account, no key rotation when you swap apps. Context caching is fully tracked: cached input tokens are metered separately so you can see the savings.

Auth modes

Profile	Default	How it works
`vendo_managed_pool` (Vendo-managed pool)	Yes	Vendo holds the upstream credential and bills per call against the tenant's Vendo balance. No tenant signup with the provider required.

Environment variables

These are the env vars Vendo injects into your deployment at boot when this integration is bound. Source values are resolved by resolveConnectionEnvVars in packages/integrations/lib/ from the connection record (credential, metadata) and deployment context (proxy URL, Vendo-issued API key).

Variable	Profiles	Source	Secret
`GEMINI_API_KEY`	all	Vendo-issued API key (`vendo_sk_*`)	Yes
`GEMINI_BASE_URL`	all	Vendo proxy base URL (`https://<slug>-proxy.vendo.run/v1`)	No

Reading the credential at runtime

Vendo writes GEMINI_API_KEY (and any companion vars above) into your deployment's environment at boot — the official Google Gemini SDK auto-discovers it, so most code just instantiates the client with no extra wiring. vendo.token(slug) is the resolution-chain-aware alternative that reads the same value in Vendo mode and falls back to whatever you set in OSS mode; use whichever fits your style. See Two modes for the full resolution chain.

Proxy endpoint

Calls to Google Gemini are brokered through Vendo's proxy at:

https://gemini-proxy.vendo.run

Point the official Google Gemini SDK at this base URL; Vendo authenticates, meters per call, and forwards upstream. Your code never sees the upstream credential.

Quickstart

import vendo

# Native Gemini API access through Vendo. Same Google Generative AI SDK, billed via your Vendo balance.
token = vendo.token("gemini")

# Use `token` wherever you would have used the provider's API key —
# pass it to the official SDK or set it as the env var the SDK reads.
print(token)

import { Vendo } from "@vendodev/sdk";

const vendo = new Vendo();
// Native Gemini API access through Vendo. Same Google Generative AI SDK, billed via your Vendo balance.
const geminiToken = await vendo.token("gemini");

// Use `geminiToken` wherever you would have used the provider's API
// key — pass it to the official SDK or read it as the env var the SDK expects.
console.log(geminiToken);

import VendoSDK

let vendo = Vendo()
// Native Gemini API access through Vendo. Same Google Generative AI SDK, billed via your Vendo balance.
let token = try await vendo.token("gemini")

// Use `token` wherever you would have used the provider's API key.
print(token)

Learn more

Concepts: connections & integrations — how connections, bindings, and credentials fit together.
Build a tool: SDK — full vendo.token and vendo.data.execute semantics.
Google Gemini API docs