What is a prompt diff?

A structured comparison between two versions of an LLM prompt. Unlike a plain text diff, it understands message structure — aligning system, user, and assistant messages by role and position, then highlighting only the content that actually changed.

Why not just use a regular text diff like git diff?

Plain text diff is structure-blind — reordering messages, renaming keys, or switching provider shapes makes every line look different despite an identical semantic prompt. Role-aware diff normalizes both sides first, then diffs matching roles to show real content changes.

Which prompt formats are supported?

The tool auto-detects and normalizes OpenAI Chat Completions, Anthropic Messages, Gemini contents, LangChain tuples, and plain text. You can even compare an OpenAI prompt against an Anthropic prompt — both are normalized to the same role-tagged message list.

Is my prompt sent anywhere?

No. All parsing, normalization, and diffing happens 100% client-side in your browser. No network requests contain your prompt data. Safe for proprietary prompts and customer logs.

What granularity should I use?

Word level is the default and best for most use cases. Use line level for large system prompts or markdown. Use character level to catch typos or invisible unicode. Use token level to see changes from the model perspective.

AI & LLM

Prompt Diff

Free online semantic prompt diff for LLM payloads. Compare OpenAI, Anthropic, Gemini and LangChain message arrays with role-aware side-by-side diff, token-level granularity, tool_use / tool_result blocks, and recursive JSON-escaped string unwrapping. Includes a raw JSON split/unified diff view and a curated gallery of 40+ prompt templates. 100% client-side.

Open tool

Quick reference

Category: AI & LLM
Runs in: Browser
Data stored: Nothing
Price: Free
Sign-up: Not needed

Browse categories

All features

Semantic, role-aware diff
OpenAI, Anthropic, Gemini, and LangChain support
Token-level granularity
tool_use / tool_result block diffing
Recursive JSON-escaped string unwrapping
Side-by-side and unified raw JSON diff
Curated prompt template gallery with 40+ templates
100% client-side, no upload

What you can do

Review prompt changes

Paste two versions of an LLM prompt to see a role-aware diff — system, user, and assistant messages aligned by position with only real content changes highlighted.

Compare across providers

Diff an OpenAI messages array against an Anthropic prompt to verify semantic equivalence after a provider migration.

Debug prompt regressions

Compare a working prompt from logs against a broken one to isolate exactly which message changed and how.

Unwrap LangChain payloads

Enable JSON-escaped unwrap to see what LangChain actually sent to the API — recursively unescaping nested JSON strings.

Choose diff granularity

Switch between line, word, character, or token-level diffing to catch everything from phrase edits to invisible unicode.

Private by design

Client-side
No sign-up
Free

Prompt Diff runs entirely in your browser tab. No data is uploaded, no server processes your input, and no account is required.

Specifications

Formats: OpenAI, Anthropic, Gemini, LangChain, plain text
Cross-provider: Yes — both sides normalized to common shape
Granularity: Line, word, character, token
Diff views: Role-aware message diff, raw split, raw unified
Filters: Ignore whitespace, ignore case, collapse blanks, unwrap JSON, wrap lines
JSON unwrap: Recursive unescaping of double-encoded strings
Templates: 40+ curated prompts across 8 categories
Processing: 100% client-side, no upload

Frequently asked questions

Related keywords

prompt diffllmopenaianthropicgeminilangchainsemantic diffrole aware difftoken levelraw diffsplit viewunified diffprompt templatesflow builderjson unwrap