Context Trimmer
Trim long text to an approximate LLM token budget while choosing whether to keep the start, end, middle, or both ends.
About this tool
Context Trimmer shortens text so it fits an approximate LLM token budget. It is useful when you need to paste a long log, transcript, prompt, article, or chat-history excerpt into a model with a fixed context window.
The tool uses a simple, transparent estimate: tokens ≈ characters ÷ characters per token. The default is 4.0 characters per token, a common English rule of thumb. Lower the value for code or non-English text when you want a more conservative trim.
Worked example
Input:
the quick brown fox jumps over the lazy dog
Settings: max_tokens = 5, chars_per_token = 4.0, keep = head_tail, marker ….
Output:
the quick…lazy dog
The marker counts toward the budget, so the final result still fits the target estimate.
Keep strategies
- Head keeps the beginning and drops the end.
- Tail keeps the end and drops the beginning.
- Middle keeps the centre and drops both ends.
- Head + tail keeps the beginning and the end, dropping the middle. Use Head ratio to split the budget between the two sides.
Limits
- This is an approximate character-based budget, not a model-specific tokenizer. Exact token counts vary by model and language.
- By default, cuts move to whitespace so words are not split. Turn on Allow cutting inside words for exact character-limit cuts.
- If the input already fits the budget, it is returned unchanged and no marker is inserted.
- The maximum budget is capped at 1,000,000 tokens to avoid accidental huge outputs.
FAQ
Does this use the same tokenizer as GPT, Claude, or Llama?
No. It intentionally uses a transparent approximation: characters divided by the Characters per token value. Use a lower value for more conservative trimming when exact model-token fit matters.
Which keep mode should I choose for logs or stack traces?
Use Head + tail for logs, stack traces, and command output. The beginning often has setup context, while the end usually contains the failure. Use Tail when only the latest messages matter.
Why does the output sometimes use fewer tokens than the target?
When word-safe trimming is enabled, the cut backs up or moves forward to a whitespace boundary so it does not split words. That can leave a little unused budget. Turn on Allow cutting inside words for stricter character use.
Is the text uploaded anywhere?
No. The trim runs locally in your browser with WebAssembly; your text stays on your device.
Developer & Automation Access
Run it from the terminal
Same engine as this page, headless — via the gizza CLI:
gizza tool context-trimmer "Paste the long text, prompt, log, transcript, or document here…"New to the CLI? Get gizza →
Open it by URL
Pre-fill and auto-run this tool with query parameters — the names match the API/CLI:
https://gizza.ai/tools/context-trimmer/?text=Paste%20the%20long%20text%2C%20prompt%2C%20log%2C%20transcript%2C%20or%20document%20here%E2%80%A6&max_tokens=512&chars_per_token=4.0&keep=head&marker=%E2%80%A6&head_ratio=0.5&break_words=trueMachine-readable descriptor: tool.json — title + parameters JSON Schema for agents.