AI & Prompts — 15 Free Online Tools

Token Counter — runs the actual provider tokenisers (gpt-tokenizer + @anthropic-ai/tokenizer in WebAssembly) in your browser. Gives you exact GPT-5/Claude/Gemini token counts and side-by-side cost estimates without a provider API key. Use it before you commit a prompt to a budget; the bespoke notes on the tool page cover BPE basics + the output-vs-input cost asymmetry that catches most teams off-guard.

Prompt Personas Library — 1,384 curated, copy-paste-ready prompts grouped into 12 categories (Development 468, Business 177, Writing 122, Productivity 64, Education 50, Creative 47, Entertainment 47, Languages 40, Lifestyle 24, Health 9, Science 9, Other 327). Search-as-you-type filter; nothing leaves the browser.

Prompt Builder / Optimizer / Templates / Personas Generator — structured editors that help you compose multi-section prompts (system + role + task + format + examples) without re-typing boilerplate. Output is a copy-pastable string; you take it to whichever model you use.

Chain-of-Thought and Few-Shot Builder — pattern scaffolds for two prompting techniques that consistently improve answer quality on reasoning + extraction tasks. The tools just generate the scaffold; you fill in the content and paste into your model of choice.

System Prompt — focused editor for the system message slot, with the common patterns (role, constraints, format hints, refusal handling) as templated sections.

Common pattern: build the prompt here (free, private, exact-token-count visible), then run it in your provider of choice (paid, cloud-resident). The split keeps the iteration cycle local.

AI Rewriter, Summarizer, Translator, Grammar Fixer, Email Writer, Code Explainer are utilities that call a language-model provider under the hood to produce their output. Unlike the prompt-engineering tools above, your input text IS sent to a model server (and from there, may be logged, cached, or used per the provider's stated policy). This is unavoidable — there's no way to summarise text without a model, and no model runs locally well enough to ship inside a browser tab in 2026.

Use them for content where that's acceptable: marketing copy, draft emails to colleagues, public documentation, code samples from open-source repos. Don't paste anything you wouldn't paste into ChatGPT directly: legal contracts, customer PII, internal Slack threads, employee names, anything covered by an NDA. The Pro paywall on these tools is partly a usage gate, partly a forcing-function that makes you stop and ask whether the input belongs in a third-party model.

Drafting a prompt from scratch → Prompt Builder if it's a multi-section system+task prompt, Prompt Personas Library if you're looking for a starting template (1,384 are likely to cover your domain).

Optimising a prompt that's working but expensive → Prompt Optimizer (rule-based shortening) + Token Counter (verify the savings + cost delta before vs after).

Your prompt isn't getting reliable answers → Few-Shot Builder (add 3-5 worked examples — usually the single highest-leverage technique) or Chain-of-Thought (force the model to reason out loud before the final answer).

Building an automated pipeline that calls an LLM repeatedly → System Prompt + Prompt Templates (extract the variable parts of your prompt into a parameterised template, lock down the system message).

No model fine-tuning interface, no embedding/vector tools. Both require infra that doesn't fit the single-tab pattern.

No 'connect your OpenAI key and chat in our UI' wrapper. Plenty of those exist (chatbox.ai, anything.io, your provider's own chat UI). They'd add maintenance surface for zero differentiation.

No AI image generation (Stable Diffusion, Midjourney, DALL-E wrappers). Different tooling tier, different audience, would muddy the category.