Analyzer

Paste anything. Count everything. Nothing leaves your browser.

Input

Paste text or drop a file. Common formats are parsed in your browser.

Redact secrets before analysis

Drop a fileor browsePDF, DOCX, XLSX, Markdown, JSON, YAML, code. Parsed in your browser.

Try a sample

Token counts

Exact counts for OpenAI and Gemini. Anthropic is a labeled estimate with an optional opt-in verify.

OpenAI

0 chars

no input

o200k_base, exact

Anthropic

0 chars

no input

estimate, cl100k_base, +/- 10%

Gemini

0 chars

no input

Gemma SPM, exact

Token visualizer

See exactly how each provider chops your text. Whitespace shows as dots; line breaks as returns.

0tokens0chars0chars / token

Nothing to tokenize yet. Paste or drop a file.

Secret redaction

runs first

API keys, tokens, and credentials are scanned and masked before any other panel sees your text.

No secrets detected.

Cost calculator

What it would cost to send this input. Output cost varies; you can model it or leave input-only.

Model

Cache hit rate

Output tokensEstimate outputrough volume assumptionCallsBatch fraction

Cache TTL

Input-token-bounded. Output cost varies by response.

Default per-call total is input-side only. Turn on the output estimate or enter an output token count to model your workload.

Audit

Rules that flag wasted tokens, with an estimated saving for each.

Experimental rules

Paste text to audit it for token waste.

Compression preview

experimental, lossy

Optional. Trim low-information words while keeping the load-bearing ones. Lossy. Off by default.

Compression preview (experimental, lossy). Downloads a ~58 MB model to your browser on first use. Runs fully locally; nothing leaves your browser.

Powered by LLMLingua-2 (TinyBERT variant). The model drops low-information words while keeping the load-bearing ones. This variant is uncased: compressed output is lowercased.

v0.1.0 scaffolding. loading core

MCP analyzer

Paste any agent's MCP config. See what its tool definitions cost and which servers to swap for a CLI.

Provider

Paste an MCP config above to see its token cost and swap recommendations.

tools/list manifest (exact counts)

Paste a server's tools/list JSON for exact per-tool counts and description smells. Tokenized with o200k_base in your browser; nothing leaves it.

Server name

Estimates are input-token-bounded. Your config never leaves the browser.