01
LLM Token Counter
Estimates how many tokens a piece of text costs, broken down by script, and shows what…
Chat Transcript Splitter
Splits a long transcript into chunks under a token budget, breaking on speaker turns rather than in the middle of one.
Developer
Nothing uploaded
No sign-up
Free, unlimited
Paste something on the left to see the result.
In short
When a transcript is too long for one call, where you cut it matters more than how big the pieces are. Splitting on a character or token count alone will cut in the middle of a speaker turn, which hands the model an answer with no question and produces chunk summaries that contradict each other. Breaking on speaker turns keeps every exchange intact. Adding an overlap of a few hundred tokens — repeating the last turns of each chunk at the start of the next — is what keeps a reference like "that approach" resolvable across the boundary, at the cost of embedding or processing a little text twice.
How to use the Chat Transcript Splitter
- Paste the transcript.
- Set a budget below the model's limit, leaving room for the reply.
- Add an overlap if later parts refer back.
- Check for turns flagged as larger than the budget.
About the Chat Transcript Splitter
The Chat Transcript Splitter splits a long transcript into chunks under a token budget, breaking on speaker turns rather than in the middle of one. It runs as plain JavaScript inside the page, which means there is no upload step, no queue and no size limit imposed by a server — the only limit is your own device.
What people use it for
- Summarising a long support conversation in pieces.
- Preparing a transcript for retrieval.
- Getting a meeting recording under a model's limit.
Details
| Breaks on | Speaker turns, falling back to blank lines |
|---|---|
| Never breaks | In the middle of a turn |
| Overlap | Repeats the tail of each chunk at the head of the next |
| Oversized turns | Reported rather than silently truncated |
Set the budget well below the model's context window rather than at it. The chunk is only the input; the instructions and the reply need room too, and a chunk sized to the window leaves nowhere for the answer to go.
Where this tool stops
Detects turns from a "Name:" pattern at the start of a line, falling back to blank lines. A transcript with no consistent speaker markers is split on paragraphs, which is looser and worth checking.
Frequently asked questions
Why not just split every N characters?
Because it cuts mid-turn. A chunk that begins halfway through an answer gives the model a reply with no question and no speaker, and its summary will confidently describe something that did not happen. That error is hard to trace, because the chunk looks perfectly readable — it is simply missing its context.
How much overlap should I use?
Enough to carry a reference across the boundary — typically 5 to 10% of the chunk. More costs proportionally more to process and adds duplicate passages to a retrieval index, which then returns near-identical results for the same query.
What if one turn is bigger than the whole budget?
It is reported rather than hidden. That turn will sit in a chunk alone and still exceed the budget, and the model will truncate it silently — so raise the budget or split that turn by paragraph first.
Do I need an account to use the Chat Transcript Splitter?
No charge and no sign-up. There is no free tier to run out of because there is no paid tier — inputtesthub.com is funded by advertising rather than by metering the tool.
Is my data uploaded anywhere?
No. The Chat Transcript Splitter runs entirely in your browser using JavaScript. Because the work happens locally, the tool also keeps working if your connection drops, and closing the tab is all it takes to discard everything.
What problem does the Chat Transcript Splitter solve?
In practice it gets used for three things: summarising a long support conversation in pieces. Preparing a transcript for retrieval. Getting a meeting recording under a model's limit. It is built for exactly those cases rather than being a general-purpose tool with this bolted on.
When is the Chat Transcript Splitter not the right tool?
Detects turns from a "Name:" pattern at the start of a line, falling back to blank lines. A transcript with no consistent speaker markers is split on paragraphs, which is looser and worth checking. Knowing where a tool stops being trustworthy matters more than the cases where it works.
Which values should I know?
Breaks on is Speaker turns, falling back to blank lines. Never breaks is In the middle of a turn. Overlap is Repeats the tail of each chunk at the head of the next. These are printed on the page itself, so you do not have to take them on trust.
Does the Chat Transcript Splitter work on a phone?
Yes. The page is responsive and the tool works in Chrome, Safari, Firefox, Edge and Samsung Internet on both Android and iOS. There is no app to install and nothing to sign into.
Related developer tools
All 9202
System Prompt Linter
Checks a system prompt for the patterns that reliably cause trouble, and shows what it costs…
03
JSON Formatter
Formats, indents and validates JSON with precise error positions, key sorting and a…
04
JSON Validator
Checks whether JSON is valid and reports the exact line, column and reason for the first error…
05
JSON Minifier
Strips all whitespace from JSON to produce the smallest valid output, showing the bytes saved.
06
JSON to CSV Converter
Converts a JSON array of objects into CSV, flattening nested keys and handling missing fields…
07
CSV to JSON Converter
Turns CSV or TSV into a JSON array of objects, with type detection for numbers and booleans.
08
JSON to XML Converter
Converts JSON into well-formed XML with a configurable root element and array item naming.
09
XML to JSON Converter
Parses XML and converts it to JSON, preserving attributes and handling repeated elements as…
10
JSON to YAML Converter
Converts JSON to clean YAML with correct indentation, quoting and multi-line string handling.