Tools
Tools for building with AI
Free tools for building with AI. Each one runs in your browser — no account, no API key, and nothing you enter is sent to a server.
Costs
- LLM cost calculatorWhat a workload costs per request, per day and per month on every model Salyro publishes, from the same prices and arithmetic Salyro prices usage with.
- LLM usage explainerPaste a usage object or a whole response from any provider and see what each count means, what is inside what, and what the request cost, slice by slice.
- Prompt caching calculatorWhat prompt caching saves on a workload — with and without the cache, the break-even hit rate, and each provider's minimum cacheable prompt.
- Batch API cost calculatorA workload sent request by request against the same workload through each provider's own Batch API, at the discount the provider states.
- AI project cost quote builderBuild the running-cost estimate a software house shows its client — features, low/expected/high volumes, reserve and margin — and print it or export it.
- Agent loop cost estimatorWhat a multi-step agent run costs, step by step — context that grows with every tool result, prompt caching on it, price bands and the context window.
- Reasoning tokens cost calculatorThe hidden cost of a reasoning model's thinking, billed as output — at the ranges you measured per effort level or budget, or from a real usage object.
Tokens and context
- Token counterCount the tokens of a text or conversation for each model — exact with OpenAI's tokenizer in your browser, a marked estimate elsewhere — with cost and context.
- Language token taxHow many more tokens the same text costs in Hebrew, Arabic, Chinese and 17 other languages than in English, measured exactly with OpenAI's tokenizers.
- Payload format optimizerConvert JSON or CSV to minified JSON, YAML, CSV, a Markdown table or XML in your browser, and see which format costs the fewest tokens.
- Context window budget plannerSplit a model's context window between system prompt, tools, history, retrieved documents, input and reserved output — with fit, price band and cost per call.
- Git diff token counterPaste a git diff or patch and see its tokens per file and per hunk, filter out lockfiles, generated and binary files, and check it fits a model for review.
Prompts
- Prompt diff with costCompare two versions of a prompt or a request's messages word by word, and see what the change does to tokens, cost per request and per month, and the cache.
- Prompt template rendererFill a prompt template's {{variables}} from JSON or a form, see missing and unused variables, and count the rendered prompt's tokens.
- Prompt linterCheck a prompt with fixed rules, no model involved — unfilled {{variables}}, contradictions, JSON without a schema, cache-breaking dates, hidden text, keys.
Structured output
- JSON Schema compatibility checkerPaste a JSON Schema and see, per provider, what is refused, what is undocumented and what changes on the way — with a version fixed for Anthropic.
- JSON repairFixes the broken JSON models return — code fences, stray text, trailing commas, single quotes, cut-off answers — with every change marked certain or a guess.
- JSON to JSON SchemaInfer a JSON Schema 2020-12 from one JSON example or several, merged — required keys, nullable values, optional enums — then validate it or make it strict.
- JSON Schema ValidatorValidate JSON against a JSON Schema 2020-12 with the validator and settings Salyro uses — every violation by its path, highlighted in your text.
- JSON Schema sample generatorBuild a sample JSON payload from a JSON Schema, then validate it with Salyro’s validator — shown as valid only when it passes, with reasons when not.
- Partial JSON parserParse JSON that has not finished arriving — a stream, tool-call arguments, a cut-off answer — into the longest valid value, each field complete or partial.
- JSON Schema to Zod, Pydantic and TypeScriptConvert JSON Schema to Zod, Pydantic v2 or TypeScript types and back, with every keyword or construct that does not convert exactly listed.
Requests and SDKs
- LLM request converterConvert a whole LLM request between OpenAI Chat, Responses, Anthropic Messages and Gemini — tools, images, JSON schemas — and see what does not carry over.
- LLM request linterCheck an LLM request against the model it is for — parameters it rejects or ignores, missing capabilities, output limits — before you send it.
- LLM parameter mappingHow each LLM request setting is spelled in OpenAI Chat, Responses, Anthropic Messages and Gemini — max_tokens, stop, tool_choice and more — in one table.
- cURL to SDK codeTurn a cURL command into OpenAI, Anthropic or Google Gen AI SDK code in Python and TypeScript, and SDK code back into cURL. The key becomes a variable.
- LLM SDK translatorTranslate one openai, anthropic or google-genai SDK call into another provider's SDK, in Python or TypeScript, with a report of what did not carry over.
- OpenAI-compatible config generatorThe base URL, model and key variable for an OpenAI-compatible endpoint, written for the OpenAI SDK, LangChain, LlamaIndex, the Vercel AI SDK and cURL.
Conversations
- Conversation linterFinds what makes a provider refuse a conversation — unanswered tool calls, orphan results, misplaced system prompts — and repairs it safely.
- Chat messages builderBuild a messages array as cards — roles, text, images, tool calls and results — and get it in OpenAI, Anthropic and Gemini formats, checked as you build.
- Conversation timeline viewerSee a long conversation as a timeline — roles, tool calls joined to their results, images, thinking, linter findings and a running token count.
- Function-calling flow simulatorDefine tools, script the calls and results, and see the exact request every provider is sent at each step — linted, compared and priced. No model runs.
Responses
- Rate-limit headers decoderPaste a response's headers and see how much is left, when each limit resets, how long to wait before retrying, and which limit decides.
- SSE stream reconstructorPaste a raw LLM stream and get the whole answer back — text, tool calls assembled from their fragments, usage and finish reason — with broken frames found.
- Finish-reason explainerWhat each provider's stop or finish reason means and what to do about it — OpenAI, Anthropic, Gemini — and the one rule Salyro applies to it.
- LLM error decoderPaste an LLM API error and see what happened, whether to retry and what to fix — for Salyro, OpenAI, Anthropic and Gemini, and what it becomes through Salyro.
MCP and coding agents
- MCP config converterConvert MCP server settings between Claude Desktop, Claude Code's .mcp.json and ~/.claude.json, and claude mcp add-json commands, with credentials moved out.
- mcp.json validatorCheck an MCP configuration for Claude Code or Claude Desktop — missing fields, bad types and names, unsafe URLs, duplicates, and credentials in the clear.
- Tool-description linterLint MCP and function-calling tool definitions — names, missing or short descriptions, undocumented parameters, provider schema rules and token cost.
- OpenAPI to MCP converterConvert an OpenAPI 3.0 or 3.1 document, JSON or YAML, into MCP tool definitions — one tool per operation, with what did not convert listed.
- MCP JSON-RPC decoderDecode MCP JSON-RPC messages — requests, notifications, results, errors — explain error codes, pair each request with its response, find the version.
- Agent instructions helperCheck and convert AGENTS.md, CLAUDE.md, SKILL.md and Copilot instruction files — frontmatter, length, secrets and token cost, in your browser.
RAG and vectors
- Document to MarkdownConvert HTML, Word (DOCX) and text-based PDF files to clean Markdown for a language model, in your browser, with tokens counted before and after.
- Chunking visualizerSee how a document splits into chunks — by characters, tokens, sentences, Markdown headings or recursively — with each chunk's tokens and the overlaps marked.
- Cosine similarity calculatorCosine similarity, dot product and Euclidean distance between vectors you paste — two, or a matrix of every pair. Runs in your browser.
- Vector storage calculatorHow much space your vectors take — by count, dimensions and precision (float32, float16, int8, binary), with metadata, index overhead and copies.
- Retrieval metrics calculatorRecall@k, precision@k, MRR and nDCG from relevance judgements and ranked results you paste — per query and averaged, with each formula stated.
- Embedding visualizerProject vectors you paste onto two dimensions with PCA and see them as a labelled scatter plot, with the points as a table too. Up to 5,000 vectors.
Images, PDFs and audio
- Image token calculatorHow many tokens an image costs on each vision model, the size the provider resizes it to, and what a batch of requests costs — computed in your browser.
- Image prep for LLMsResize, convert and compress an image in your browser to the size a model sees, with tokens and bytes before and after, and its EXIF and GPS removed.
- PDF token and cost estimatorEstimate a PDF's tokens and cost on each model whose provider documents PDF input, and check its page and size limits — read in your browser.
- Audio prep and cost calculatorRead a recording's length, convert it to mono WAV at the sample rate you choose and split it by size — in your browser — with what providers charge for audio.
- Image to Base64 and data URI converterTurn an image into a data URI, base64 or the content part each provider format expects, and a data URI back into an image — in your browser.
Data preparation
- Batch JSONL builderTurn a CSV, JSON or JSON Lines file and one request template into Anthropic Message Batches requests, checked, split at the limits and priced.
- Batch results mergerJoin an Anthropic Message Batches results file back to its rows by custom_id, with failures, gaps and duplicates flagged and the actual cost.
- Fine-tuning dataset validatorCheck a fine-tuning JSONL file against OpenAI, Gemini on Vertex AI or Claude in Amazon Bedrock rules, line by line, and count its tokens.
- Dataset profilerProfile a JSONL or CSV dataset in your browser — token lengths, scripts, repeated rows, missing fields, outliers and a seeded random sample.
- Eval rubric builderBuild an evaluation rubric — criteria, scored levels, an example per score — and export it as a judge JSON Schema, a judge prompt, or CSV for labelling.
- Judge agreement calculatorMeasure how far raters agree — people, LLM judges or both — with percent agreement, Cohen's and weighted kappa, Fleiss' kappa and Krippendorff's alpha.
Privacy and security
- PII & secret redactorReplace API keys, tokens, emails, phones, card numbers, IPs and IBANs in a log or prompt with consistent placeholders, in your browser, before sharing.
- Invisible text & homoglyph detectorFind zero-width characters, hidden Unicode tag text, bidi overrides and lookalike letters in a prompt or document, decode them, and clean them in one click.
- API-key leak scannerScan code, .env files, configs and logs in your browser for AI-provider keys, Salyro keys and cloud tokens, by file and line, values masked, with next steps.
- MCP permission auditSee which MCP tools write, delete, run commands or reach the network — server annotations kept apart from guesses — with explicit permission rules.
AI economics
- AI unit economics calculatorWhat the AI in your product costs per user, per customer and per feature each month — as a share of revenue, a gross margin and a break-even point.
- LLM throughput plannerThe requests and tokens per minute your product needs at its busiest, how many requests are in flight at once, and whether your own rate limits carry it.
- Model switch impactWhat moving a workload from one model to another changes — cost at the same load, capabilities, rejected or ignored parameters, context window and lifecycle.