The output
What the Llama tone analyzer reports
A short written reading of the passage you pasted: how it is likely to come across, which parts carry that impression, and where a reader could take it more than one way. Ambiguity is often the useful part, because a line you thought was neutral and a reader could take as a deadline is exactly the line worth changing.
Nothing is scored and nothing is rewritten. Tools exist that adjust tone for you, sliding a message from direct to friendly. This is not one of them, which means the words that go out are still the words you chose.
The use case
Checking a message before you send it
Most people arrive here holding one message and a bad feeling about it. The check takes a few seconds, and what you paste changes the answer more than anything else you do.
A reply read on its own is missing half its meaning. Paste the context when the context is doing work.
- Paste the thread
- Include the message you are answering when your tone is a response to theirs. Curt after a complaint reads differently than curt out of nowhere.
- Paste the whole message
- Openings and sign-offs carry a lot of the impression. A body paragraph alone will be read colder than the real email.
- Leave out the signature block
- Legal footers and disclaimers add nothing and use up characters.
Workflow
How to run the Llama tone analyzer in three steps
Your text is posted to our AI server, read there and returned in the same response. No account is created at any point.
- Step 1
- Paste the message, 1 to 12,000 characters. Anything longer is refused before the check is spent.
- Step 2
- Name the reader and the effect you were going for in up to 300 characters, then run it.
- Step 3
- Change the one or two lines the findings point at, and re-check only if the rewrite was substantial.
Setup
Telling the analyzer who the reader is
Without a reader in mind, a tone reading has to hedge across every possible audience, and hedged findings are the ones that feel obvious. The instruction field is where you remove that hedge. Other tools sell audience presets and formality sliders for the same purpose. Here it is 300 characters of your own words, which fits more situations than a dropdown does.
Name the relationship and the outcome you want, not just the register.
- Useful
- Going to a client who missed two deadlines. Should read as firm about the date and not personal.
- Useful
- Internal note to someone more senior. I want direct, not blunt, and no apologising for asking.
- Weak
- Make sure the tone is good. There is no reader and no target in that, so the reading comes back generic.
Limits
Five checks a day, and what to spend them on
Five per rolling 24 hours, and the window moves forward from your first check rather than resetting overnight. That is enough for the messages that genuinely matter and not enough to check every email you write, so the triage is the skill.
Spend them on messages where the relationship is at stake or the reader is not obliged to be generous: a complaint response, a first approach, feedback to someone who did not ask for it, a message written while annoyed. Routine confirmations do not need a check, and re-running the same message after a one word edit is the fastest way to waste the allowance.
Scope
What the Llama tone analyzer will not do
It will not tell you what the other person meant. Running someone else's message through it produces a description of their words, not access to their intent, and treating the output as evidence about a colleague's feelings is a misuse of it.
It will not give you a number, a per sentence breakdown, an emotion label from a list, or a rewritten version. It will not detect sarcasm reliably, because sarcasm depends on history between two people that is not in the text. And it is a check rather than an approval: tone is decided by the reader, and no tool can stand in for one.
Privacy
A server route, and messages you should not paste
Reading tone needs a language model, so this text goes to our AI server. Some tone checkers run entirely inside the browser page and transmit nothing, which is a genuine advantage for private correspondence, and pretending otherwise would be dishonest about how this route works.
Practical rule: do not paste a message containing another person's private information, HR matters, medical details or anything under contract. Paraphrase the structure of the message instead, or check a version with the specifics removed. The reading is about the shape of the writing, and the shape survives paraphrase.
In practice
Who runs the Llama tone analyzer, and when
Support teams answering complaints where the first sentence decides how the rest is read. Founders sending a first cold message and unable to tell whether it reads as confident or pushy. Anyone writing across a language or a culture gap, where formality is easy to overshoot in both directions.
The shared moment is the pause before sending. You have reread it until the words stopped meaning anything. This gives you an outside reading in a few seconds, then hands the decision straight back.