Step by step
From issuing a key to your first chat
You only do this once. If you already use a provider, paste that key — keys are always issued in the provider's own console.
Before you start
You need one provider account
- A provider account
- Credits or a free quota
- The browser you will paste into
1. Provider
Where to create the key
Using a provider you already pay for is the fastest path. Every link below goes to the provider's official console.
Gemini (Google AI Studio)
Free tierUse Gemini within the free tier of Google AI Studio.
Key formatAIza...
Browser calls work. AI Studio keys are meant for prototyping, so restrict the key.
Claude (Anthropic)
Prepaid creditsCalls Claude models directly from your browser with your own key.
Key formatsk-ant-...
Direct browser calls are supported. KeyChat sends the required opt-in header.
ChatGPT (OpenAI)
Prepaid creditsCalls GPT and o-series models directly. Billing must be set up on your account.
Key formatsk-...
Not the officially recommended path. If CORS blocks it, route through OpenRouter.
DeepSeek
Low-cost prepaidCalls low-cost DeepSeek models directly. Requires prepaid credit on your account.
Key formatsk-...
Direct browser calls (CORS) work. Connected through the OpenAI-compatible API.
OpenRouter (multi-model)
Some free modelsOne key for almost every model. The most reliable choice for direct browser calls.
Key formatsk-or-v1-...
Officially supports direct browser calls — the smoothest option.
2. Steps
Five steps, start to finish
Billing and quota setup differ by provider. This order avoids most dead ends.
- 01
Create a provider account
Skip this if you have one. OpenAI and Anthropic need a payment method; Google AI Studio only needs a Google account.
- 02
Create an API key
Use the API Keys section of the console. Most providers show the key only once — copy it right away.
- 03
Check billing and quota
OpenAI and Anthropic require prepaid credits. Google AI Studio works within a free quota, and OpenRouter needs credit added first.
- 04
Paste it into KeyChat
Open Settings from the chat screen, pick the provider and model, paste the key and save. Choose whether it is remembered in this browser, until the tab closes, or forgotten on reload.
- 05
Test the connection, then chat
Use Test connection in Settings to confirm the key works, then send your first message. If it fails, check Troubleshooting below.
KeyChat
AI character chat on your own API key. No chat fee — funded by ads and optional purchases.
3. Cost
What it will cost
The provider sets the price and KeyChat adds nothing. The numbers below are estimates from public list prices.
Claude Sonnet 5.5
Anthropic
Input / 1M $2.000 · Output / 1M $10.000 · One chat $0.012
$2.000
$10.000
$0.012
Claude Haiku 5.5
Anthropic
Input / 1M $0.100 · Output / 1M $0.500 · One chat $0.0006
$0.100
$0.500
$0.0006
GPT-6.1 Sol
OpenAI
Input / 1M $2.000 · Output / 1M $10.000 · One chat $0.012
$2.000
$10.000
$0.012
GPT-6 Luna
OpenAI
Input / 1M $0.100 · Output / 1M $0.500 · One chat $0.0006
$0.100
$0.500
$0.0006
Gemini 3.8 Flash
Google
Input / 1M $1.500 · Output / 1M $7.500 · One chat $0.009
$1.500
$7.500
$0.009
Gemini 3.5 Flash-Lite
Google
Input / 1M $0.300 · Output / 1M $2.500 · One chat $0.003
$0.300
$2.500
$0.003
DeepSeek Flash
DeepSeek
Input / 1M $0.300 · Output / 1M $1.200 · One chat $0.002
$0.300
$1.200
$0.002
1,000 input + 1,000 output tokens
Providers can change prices at any time. Your actual bill depends on provider policy and caching or batch discounts — check the provider's pricing page before relying on a number.
The cheapest start: use the Gemini 3.8 Flash free quota and only reach for paid models when you need them.
4. Browser calls
Your browser talks to the provider directly
KeyChat has no server, so requests go from this browser straight to the provider API. If a provider does not allow browser calls, CORS can block the request.
Google AI Studio
Supported
Browser calls work. AI Studio keys are meant for prototyping, so restrict the key.
Anthropic
Supported
Direct browser calls are supported. KeyChat sends the required opt-in header.
OpenAI
Conditional
Not the officially recommended path. If CORS blocks it, route through OpenRouter.
DeepSeek
Supported
Direct browser calls (CORS) work. Connected through the OpenAI-compatible API.
OpenRouter
Supported
Officially supports direct browser calls — the smoothest option.
Keys live in browser storage. Use a device you trust, and switch the storage mode to 'Until this tab closes' if your browser has a lot of extensions.
5. Local models
Models running on your own machine
Any OpenAI-compatible endpoint works. Nothing has to leave your computer.
Ollama example
# 1. Pull a model
ollama pull llama3.2
# 2. Allow browser requests (macOS · Linux)
OLLAMA_ORIGINS=* ollama serve
# Windows (PowerShell)
$env:OLLAMA_ORIGINS="*"; ollama serveOllama blocks direct browser requests by default. Setting OLLAMA_ORIGINS allows them.
KeyChat settings
- Provider
- Custom (OpenAI compatible)
- Endpoint
- http://localhost:11434/v1
- Model
- llama3.2 (the model you pulled)
- API key
- Optional — leave empty
LM Studio uses http://localhost:1234/v1, Groq uses https://api.groq.com/openai/v1.
6. Troubleshooting
Where people get stuck
Most failures come from key format or browser call restrictions.
CORS error · Failed to fetch
The provider blocks browser calls, or an extension or corporate proxy interfered. Switching to an OpenRouter key usually fixes it.
401 · authentication failed
Check for spaces around the key, a truncated paste, or the wrong provider selected. Anthropic keys only work with Claude.
429 · rate limited
You hit a free quota or per-minute limit. Wait a moment, or check billing status in the provider console.
400 · model not found
Check the model id. OpenRouter uses provider/model ids. If a model is missing from the list, type its id directly.
Add your key in Settings and start
The key is stored only in this browser and can be deleted at any time.