Ellavox AI

Automatically detect prompts in your files and show token counts with cost estimates for GPT-4o, GPT-4, Claude, and Llama 3.

Features

Auto-detect prompts in any file type:
- Markdown — headers (## System Prompt), fenced code blocks, XML tags (<system>...</system>)
- Code files — variables named prompt, systemPrompt, userMessage, etc. in JS/TS/Python
- Plaintext / .prompt — treats the entire file as a prompt
Token counts via CodeLens annotations above each detected block
Cost estimates based on current model pricing (per 1M input tokens)
Hover details — hover over any prompt block for a full breakdown: tokens, characters, words, lines, cost, and context window usage
Model switcher — click the status bar or any CodeLens to switch between tokenizers
Count selection — select any text and run the command to count tokens

Model	Tokenizer	Price (input/1M)	Context Window
GPT-4o / o1 / o3	o200k_base	$2.50	128K
GPT-4 / GPT-3.5	cl100k_base	$30.00	128K
Claude	cl100k_base (approx.)	$3.00	200K
Llama 3 / 3.1+	Llama 3 BPE	$0.59	128K

Claude token counts are approximate — Anthropic does not publish a standalone tokenizer for Claude 3+.

Open any file containing prompts
Token counts appear as CodeLens above each detected prompt block
Hover over a prompt block for a detailed breakdown
Click the status bar item (bottom-right) or any CodeLens to switch models
Select text and run Ellavox AI: Count Tokens in Selection from the command palette

Command	Description
`Ellavox AI: Select Model`	Switch the active tokenizer model
`Ellavox AI: Count Tokens in Selection`	Count tokens in the current text selection

Setting	Default	Description
`llmTokenCount.defaultModel`	`gpt-4o`	Default tokenizer model

In Markdown / plaintext:

Sections under prompt-related headers (## System, ## User, ## Prompt, etc.)
Fenced code blocks
XML-style tags (<system>, <user>, <assistant>, <prompt>, etc.)

In code files (JS/TS/Python/etc.):

Variables whose name contains prompt, system, user, assistant, message, instruction, or context
Supports template literals, triple-quoted strings, and single-line strings

npm install
npm run build    # one-time build
npm run watch    # rebuild on changes

Press F5 to launch the Extension Development Host for testing.

MIT