Model Access

API Keys Setup Guide

Set up provider API keys and check project permissions, model access, and billing before connecting your research tools.

Beginner · 20–30 min · Provider account or assigned key; Cherry Studio for configuration

Models are the engines. API keys are the ignition switches that let you choose which to start.

Overview

Use this guide when your chosen tool needs a provider API key. Start with access you already have:

Course supplied a key? Use its assigned provider in Cherry Studio. You do not need to create a Google Cloud project.
Already using Google Cloud? Follow Vertex AI setup.
Arranging access for a class? Start with Model access for a class.

The recipes below cover Google AI Studio, OpenRouter and direct provider keys. They are alternatives, not accounts everyone must open.

Benefits:

Access to AI models for different tasks
Free tier options for testing and development
Redundancy across providers for reliability
Cost-effective access through multiple free tiers

API-key routes:

Google AI Studio for the direct Gemini API recipe below
OpenRouter for testing models through one gateway

Optional providers:

Anthropic Claude (via API)
OpenAI GPT models
Other specialized providers

Provider setup

Direct Gemini API access

Google AI Studio gives you direct Gemini API access with free-tier options. Use it for course work, testing, and backup access when other providers are unavailable.

Get your API key:

Sign in to Google AI Studio (opens in a new tab) and accept its terms.
Open Dashboard, then Projects. New users may receive a default project and key. If you already use Google Cloud, import the project you intend to use.
Open API Keys and create a key in that project. New AI Studio keys are authorization keys bound to a service account.
If you cannot create a key, ask the project administrator for the required IAM permissions (opens in a new tab). These cover project access, key creation, API enablement, and service-account creation and binding.
Store the key securely. Keep its Generative Language API restriction. Google rejects unrestricted standard keys. Follow the key management guide (opens in a new tab) for older keys and application restrictions.

Free access covers selected models and quotas, not every Gemini model. Check the project's tier in AI Studio. Paid access requires an active linked billing account. If Google assigns your project to Prepay, fund that balance too. Follow Google's billing setup (opens in a new tab) before selecting paid models.

For Google SDKs, set GEMINI_API_KEY or GOOGLE_API_KEY. If both exist, the SDK uses GOOGLE_API_KEY. A client with its own key field may behave differently. Check which key and project your tool actually uses.

Current Gemini API models:

Model	Status	Best for
Gemini 4 Argon	Stable, paid only	New flagship candidate; check client support before switching
Gemini 3.8 Flash	Stable	Starting choice for sustained coding, agentic, and research tasks
Gemini 3.1 Pro	Preview	Complex reasoning and large-context analysis
Gemini 3 Flash	Preview	Frontier-class performance at lower cost
Gemini 3.5 Flash-Lite	Stable	High-volume summarization, tagging, and screening
Gemini embedding models	Varies	Document similarity and semantic search

Argon has no Free tier. Its provider availability does not establish support in Cherry Studio or another client; check that client's API route and controls first. See the model reference for context, pricing, and migration limits. Documentation-confirmed on October 19, 2026; account API access not tested.

Free-tier limits change frequently. Check the official rate-limits page (opens in a new tab) and Gemini model list (opens in a new tab) before relying on a quota for teaching or batch work. Google also offers student memberships and promotions; check Google's current student offers (opens in a new tab) before paying. Eligibility, plan and renewal terms vary. These memberships are separate from model-API billing and Google Cloud teaching credits.

Configure in Cherry Studio:

Open Cherry Studio → Settings → Model Services
Add Google AI Studio
Paste your API key
Choose Gemini 3.8 Flash as the default, or Gemini 3.1 Pro preview for complex analysis
Test the connection

Recommended - Multi-Model Access

OpenRouter gives you access to 100+ AI models through a single API. Test different models without managing multiple accounts.

Setup steps:

Visit OpenRouter.ai (opens in a new tab)
Sign up with email or Google account
Add credits (minimum $5 recommended):
- Use credit card or crypto
- OpenRouter may expire unused credits after one year; check its current credit terms (opens in a new tab)
- Pay once, use across all models

Get your API key:

Go to Keys section (opens in a new tab)
Click "Create Key"
Name it "Systematic Review Course"
Copy the generated API key

Warning

OpenRouter charges a service fee when you buy credits.

Configure in Cherry Studio:

In Cherry Studio Settings → Model Services
Add OpenRouter
Paste your API key
Start with Claude Sonnet 5.5 (opens in a new tab) from OpenRouter's catalog.
Test multiple models by changing the model name

Benefits:

100+ models from various providers
Free models for testing available
Single API for all providers

For a course or a lab: one account, several keys. Buy credits once, then create a named key per participant and use the Set a credit limit option on it. Each key shows its own usage. Set a lifetime cap or choose a reset interval deliberately. OpenRouter credit limits exclude bring-your-own-key (BYOK) spend by default; enable include_byok_in_limit to count it. Follow the current cap documentation (opens in a new tab) and check the configured scope before distributing keys. For larger cohorts the provisioning API (opens in a new tab) creates and disables keys programmatically. This is the arrangement the systematic review case study uses. OpenRouter is one gateway of this kind; Haimaker (opens in a new tab) is another. Vercel AI Gateway (opens in a new tab) keys can each carry a budget of their own, and Cloudflare AI Gateway (opens in a new tab) routes requests through your own provider keys or a prepaid balance. Whichever you choose, test buying credits before the course starts.

Subscriptions you already hold

Model retirement can differ by access route. GPT-5.5 retired from Codex with ChatGPT sign-in on October 14, 2026; the official notice (opens in a new tab) leaves the OpenAI API and API-key-authenticated Codex sessions unaffected. The API example below is not a subscription-menu recommendation.

A ChatGPT subscription can include Codex usage, and lighter models can stretch its included allowance further. Access and limits depend on your plan and workspace. The local Codex app and CLI also support custom providers through user-level configuration, including OpenRouter's documented integration (opens in a new tab). Check the current pricing (opens in a new tab) and your usage dashboard: included usage, rate limits, and additional credit purchases are different things.

One local endpoint for all of them (advanced)

CLIProxyAPI (opens in a new tab) wraps the logins behind Antigravity, Codex, Claude Code, Grok Build, and Kimi into one local server that speaks the OpenAI, Gemini, and Claude API shapes. Clients see a single base URL, http://127.0.0.1:8317 by default, and a single key, so a Cherry Studio, Codex, or Claude Code session can reach supported models exposed by its configured accounts and routes, subject to their access limits. It does not make every subscription model available. The author documented using this route for Kimi and Grok inside Codex in October 2026; this section does not describe the current local configuration.

# macOS, via Homebrew
brew install cliproxyapi
brew services start cliproxyapi

Other platforms use the project's installer or its desktop client, EasyCLIProxyAPI (opens in a new tab); the project documentation (opens in a new tab) has the client configuration for each app. Two cautions. Setting it up is fiddly, which makes it a good first job for the agent itself. Whether routing a consumer subscription through an API is allowed depends on that provider's terms. Read the applicable subscription terms before building a course on this route.

Testing your setup

Most platforms include built-in testing tools:

Cherry Studio:

Select several models in the input area to compare responses side by side. You can also mention them, for example: @gpt-5.5 @claude-sonnet-5-5 Compare these approaches....
See Cherry Studio Setup Guide for detailed testing instructions

OpenRouter:

Visit openrouter.ai/chat (opens in a new tab) for web-based model testing
Try free models without setup
Compare responses across different providers
OpenRouter charges a service fee when you buy credits.

Cost management

Free and paid tiers:

Google AI Studio: Selected models have a free tier, subject to project eligibility and quotas
OpenRouter: Free models for testing, pay-as-you-go for extensive use (service fee applies)
Other providers: Free and paid options available

Budget tips:

Start with free tiers for testing
Monitor usage in provider dashboards
Use course-provided API budget for intensive work

Security best practices

API key storage:

Use password managers or secure notes
Never store in plain text or shared documents
Protect Cherry Studio's local configuration and backups; revoke and replace any exposed key

Multiple providers:

Set up Google AI Studio if its selected model and project qualify for free access
Use OpenRouter for testing different models
Have backup providers ready

Quick reference

Provider comparison

Provider	Setup Difficulty	Free Tier	Best For
Google AI Studio	Project-dependent	Selected models and quotas	Direct Gemini access
OpenRouter	Easy	Free + Paid	Model testing/comparison
Other Direct APIs	Medium	Varies	Specialized model access

Recommended configurations

For beginners:

Use a course-assigned provider or access you already have
Add another provider only when your task needs it

For advanced users:

Google AI Studio for eligible free-tier work or deliberately funded paid models
OpenRouter (model testing)
Additional providers as needed

For budget-conscious users:

Eligible free-tier models, with their quotas checked
OpenRouter (cost-effective multi-model)

Cost estimates

Choose a spending limit you can afford, then run a small representative task. Measure input, output, reasoning, and tool charges before estimating a batch. Repeated long-context requests can cost much more than short chats; a monthly price without a workload is not a useful estimate.

Resources

cite this page

Lin, X. (2026). API Keys Setup Guide. Research Memex. https://research-memex.org/docs/implementation/ai-environment-setup/api-keys-setup-guide

@misc{docs-implementation-ai-environment-setup-api-keys-setup-guide-2026,
  author = {Xule Lin},
  title = {API Keys Setup Guide},
  year = {2026},
  howpublished = {\url{https://research-memex.org/docs/implementation/ai-environment-setup/api-keys-setup-guide}},
  note = {ORCID: 0000-0001-7885-4194}
}

one renderingthe source remains