AI tools: fix AI output, count tokens, check tool schemas and agent runs

Fix and clean AI output

• JSON Repair: Paste almost-JSON from a chat reply, a log or a config. Get strict JSON back and a list of every fix, with line and column.
• Remove Markdown: Strip asterisks, hashes, links and tables from ChatGPT or Markdown text. Choose how bullets, links and tables come out.
• Invisible Characters: Find hidden characters in pasted text, see each one as a code point, and remove or replace them by type.
• HTML to Markdown: Paste HTML and get clean Markdown: headings, links, lists, code blocks and tables. Scripts and styles are dropped.
• Log Condenser: Strip colour codes and timestamps, collapse repeated lines and keep the errors. See how much shorter the log got.
• Redact Secrets: Paste logs, config or a prompt and replace API keys, tokens, passwords, emails and card numbers before you share it. Nothing is uploaded.

Tokens and cost

• Token Counter: Paste text to get the exact token count for OpenAI's o200k_base encoding, see where it is cut, and check it against context sizes.
• LLM Cost Calculator: Type your tokens per request, request volume and per-1M-token prices to get cost per request, day, month and year. Nothing is uploaded.
• Text Chunker: Split text by characters, words, approximate tokens, Markdown headings or paragraphs, with overlap. Copy the chunks or a JSON array. Nothing is uploaded.
• Prompt Cache Analyzer: Paste two API request bodies. See the first block that changed, the shared prefix size and whether it meets the minimum cacheable length.
• JSON Summarizer: Paste a large API response. Long strings and lists are cut with markers, nulls and empties are dropped, and you see tokens before and after plus an outline of the structure.

Schemas and tool definitions

• JSON to JSON Schema: Paste a JSON example or a list of samples and get a draft 2020-12 schema with types, required keys and formats. Nothing is uploaded.
• JSON Schema Validator: Paste a schema and some JSON and see every error with its path, keyword and message. Draft 2020-12 core rules. Nothing is uploaded.
• Schema Strict Mode: Paste a JSON Schema. Get a version the chosen API accepts, a list of every change by JSON Pointer path, and warnings for rules that cannot be kept.
• AI Tool Converter: Paste one tool or a list in any of five formats. The format is detected, converted to the one you pick, and checked against that vendor's name and schema rules.
• OpenAPI to AI Tools: Paste an OpenAPI 3.0 or 3.1 spec as JSON or YAML. Each operation becomes one tool with merged parameters, resolved references and a clear list of what was left out.
• JSONL Validator: Check every line of a JSONL file, turn JSON Lines into a JSON array or a JSON array into JSONL, and check the chat fine-tuning format. Nothing is uploaded.
• Tool Schema Linter: Paste tools from OpenAI, Anthropic, Gemini or MCP, even from several servers. Get name errors, weak descriptions, schema mistakes, name collisions and a token estimate per tool.
• JSON Schema to Zod: Paste a JSON Schema (draft-07 or 2020-12) and get Zod 4 or Zod 3 code, optional TypeScript types, and a table of the keywords that could not be converted.

Agent setup

• MCP Config Generator: Enter a server once. Get the config entry for your client, the file path for macOS, Windows and Linux, and a warning if a secret would end up in plain text.
• AGENTS.md Generator: Fill in your project once. Get AGENTS.md, CLAUDE.md, a Cursor .mdc rule with frontmatter and a SKILL.md skeleton, each ready to copy, plus lint notes from the vendors' docs.
• Unified Diff Checker: Paste a unified diff to check its hunk headers and line prefixes, get a corrected diff and apply it to your file. Nothing is uploaded.
• Context Pack Builder: Add files, a folder or a .zip and get one Markdown, XML or plain-text pack with a file tree and a token estimate. Nothing leaves your browser.
• SKILL.md Validator: Paste an agent skill, a Cursor .mdc rule or an AGENTS.md. Get errors and warnings with line numbers, checked against the published rules.

Your website and AI crawlers

• llms.txt Generator: Build an llms.txt file from a list of pages, or paste one to check its H1, links and sections. Follows the llmstxt.org proposal. Nothing is uploaded.
• Block AI Crawlers: Pick training only, all AI bots or your own list and copy the robots.txt lines, with each bot's purpose from its operator's own page where there is one. Nothing is uploaded.

Check prompts and agent runs

• Prompt Template Tester: Paste a prompt with {{var}}, {var} or ${var} placeholders and your values. See the rendered prompt, missing and unused variables, and a token estimate.
• Tool Call Validator: Paste an Anthropic, OpenAI Chat or OpenAI Responses conversation. Get a list of broken tool_use, tool_calls and result pairs with the message index.
• Exact Math Calculator: Type one expression per line. Get exact fractions, mixed numbers and decimals with repeating digits marked, and no floating-point error.
• Apply Patch: Paste a unified diff and the original text. Get the patched text and a status for every hunk: exact, offset, fuzz or failed, with the reason.

Safety checks for agents

• Prompt Injection Scanner: Paste web page text, tool output or a document. See instruction overrides, hidden characters, data-leaking links and encoded text, with line and column.
• Shell Command Checker: Paste a bash or zsh command from a chat, a README or an AI agent. See what each part does and which parts are risky, with a reason for each.
• Shell Escape: Quote one string or a list of arguments for five shells, and see how each shell splits the result, so you know it reads back unchanged.

Embeddings and retrieval

• Cosine Similarity: Paste two vectors, a list of embeddings or two texts. Get cosine similarity, angle, dot product, euclidean distance and norms, or a ranked list.
• Retrieval Evaluation: Paste ranked results and the relevant ids for each query. Get precision@k, recall@k, MRR, MAP, nDCG@k and hit rate per query, or compare two runs side by side.

About these tools

These tools work on the text, JSON and files that AI models produce and read. They are for people who paste ChatGPT, Claude or Gemini answers into other places, for developers who call model APIs, and for AI agents that drive a browser and need a step done exactly. Each one runs in your browser tab, so nothing you paste is sent to a server or to any AI model.

Pick the tool by the job. If a model gave you broken JSON, JSON Repair fixes code fences, trailing commas, single quotes and cut-off endings and lists every change. Remove Markdown turns asterisks and hash signs into plain text, Invisible Character Detector finds zero-width and other hidden characters, HTML to Markdown turns a web page into text a model can read, and Log Condenser shrinks a long log before you paste it. Redact Secrets masks API keys, tokens and email addresses before you share text. For budgets, Token Counter counts tokens exactly for OpenAI's o200k_base encoding, LLM API Cost Calculator turns token counts and the prices you type into a monthly cost, Text Chunker splits long text for retrieval, Prompt Cache Analyzer shows the first block that changed between two requests, and JSON Summarizer shortens a large API response.

For structured output and tool calling, the schema tools infer, validate and convert JSON Schema. Tool Schema Linter checks tools from OpenAI, Anthropic, Gemini or MCP for name errors, weak descriptions and schema mistakes, AI Tool Converter converts a tool between five formats, OpenAPI to AI Tools turns an OpenAPI file into tool definitions, and JSON Schema to Zod writes Zod 4 or Zod 3 code. To check a run, Tool Call Validator finds broken tool_use, tool_calls and result pairs, Prompt Template Tester renders a prompt with your values, and Prompt Injection Scanner and Shell Command Checker flag risky text and commands before an agent acts on them. Cosine Similarity and Retrieval Evaluation score embeddings and ranked results.

Agent setup tools write MCP config files, AGENTS.md or CLAUDE.md files and SKILL.md skeletons, check skills and rules, pack files into one prompt, and check or apply unified diffs. For your own website, llms.txt Generator and Block AI Crawlers write the files that tell AI systems what to read.

Frequently asked questions

Is anything I paste sent to ChatGPT, Claude or another AI model?

No. These tools are plain code that runs in your browser tab. A 2 MB log in Log Condenser or a JSON file in JSON Repair is processed on your device, and no request carries its content. The pages count visits with Google Analytics in rough groups, never the text.

Can an AI agent use these tools?

Yes, an agent that drives a real browser can open a page such as /json-repair, paste text into the input box and read the result box, which updates as the text changes. The results are deterministic: the same 2 KB of broken JSON always gives the same repaired output and the same list of fixes, which is useful when an agent needs an exact count or a strict format.

Is the token count the same for Claude and Gemini?

No. Token Counter is exact for OpenAI's o200k_base encoding. Anthropic and Google use their own tokenizers, which are not published as a library, so the same text gives a different number there. Anthropic offers a token counting API for exact Claude counts.

Why are vendor prices and bot names marked with a date?

Model prices, tool-definition formats and crawler names change. The cost calculator asks you to type current prices instead of shipping a price list, and Block AI Crawlers and the schema tools say which vendor documentation they follow, checked in October 2026.

Does a clean result from Prompt Injection Scanner mean the text is safe?

No. It matches fixed patterns only, such as instruction overrides, hidden Unicode tag characters (U+E0000 to U+E007F) and Markdown images whose address carries data. A new phrasing, a non-English instruction or a request spread over three sentences can pass. Keep a person or a permission prompt in front of any action.

What does Tool Call Validator find?

Paste a conversation in one of 3 formats, Anthropic, OpenAI Chat or OpenAI Responses, and it lists broken tool_use, tool_calls and result pairs with the message index.