Get started
One click. It's yours right away, and nobody has to send it to you.
Paste your key, give Vesper a character to play, and chat.
Copy a few lines of Python, JavaScript or Godot.
Models
Prices are per million reply tokens and shown on each model. You only pay for the reply, not your prompt.
Playground
API keys
Your key lets your app use Vesper. Send it with every request, and keep it secret, like a password.
How to use your key
Put the key in the Authorization header of each request. With an OpenAI library, pass it as the API key and change the base URL:
Usage
What your API key used, per day.
Billing
Credit pays for API replies. When it runs out, requests stop until you add more.
Quickstart
Your first Vesper reply in about two minutes.
1. Get a key
Open API keys and click Create API key. Copy it somewhere safe.
2. Send a request
Vesper works like OpenAI's Chat Completions API. If you already use an OpenAI library, change two things: the base URL and the key.
3. Make it yours
- Put a system message first to give Vesper a role. For example: "You are a shopkeeper. Answer in one or two sentences."
- Send the whole conversation each time (your messages and Vesper's replies) so it remembers what was said.
- Add
"stream": trueto get the reply word by word, like the chat site does.
Keep your key on your own server or inside your game. Don't put it in a public web page, where anyone can copy it.
Vesper for games
NPCs that actually talk back. Short replies are cheap: about 8,000 NPC lines for $1. Want them offline, inside the game? Download a model (coming with the game pack).
The idea
Give each character a system message with who they are, what they know and how they talk. Send the player's line and show Vesper's reply. Keep reasoning_effort: "minimal" and a small max_tokens for fast, short answers.
Tips
- Keep system messages short and concrete: name, job, mood, two or three facts about the world.
- Keep only the last few lines of each conversation, so prompts stay small and replies stay fast.
- Vesper is a small model, so check what it says. Don't let it decide things that break your game, like prices, quests or items.
- For a released game, send requests through your own small server, so players can't pull your key out of the game files.
API reference
Base URL https://vesper-chat.com/v1, sent with the header Authorization: Bearer YOUR_KEY.
Endpoints
The models your key can use. "default": true marks the default model.
Writes a reply to a conversation.
Request body
A model id, such as sonata-1 or motif-2, or default.
A list of {role, content} with roles system, user and assistant. The last one must be user. Only text is supported.
true streams server-sent events that end with data: [DONE]. stream_options: {"include_usage": true} adds token counts at the end.
The longest reply you want, up to 600.
From 0.1 to 1.0. It can lower Vesper's usual setting but not raise it.
minimal, low, medium (the default), high, extreme or max. Every mode allows the same reply length. On code and math, higher modes write more drafts and keep the one whose code runs or whose math checks out. They're slower, and every draft is billed (up to 2×, 4×, 8× or 16× on code and math; other requests are one reply in every mode).
Response
content is the reply. reasoning_content has the model's thinking steps, when there are any (only in replies that aren't streamed).
stop, or length when the reply hit max_tokens.
Prompt and reply token counts. Only reply tokens are charged, including extra drafts on code and math.
Extras: tools lists the calculator and Wikipedia lookups the model used, and sources lists the articles it read.
Errors
Missing or wrong key.
Out of credit, or more than 20 requests a minute.
The model is busy, so try again in a moment.
Vesper is a small model trained from scratch, and it runs on a home PC. It makes mistakes and can be slow or offline at times. The terms and privacy rules apply.