Read local text
Paste or upload Markdown and plain-text agent files. The current alpha limit is 1000 characters for anonymous use.
Context efficiency for agents
Measure agent-context bloat before it hits the model.
CheapAgent flags duplicate and vague instructions, measures before/after size, and converts agent docs to TOON when structured output is a fit. It runs in your browser and does not claim savings it has not measured.
Profile first
detect structure
Evidence receipt
reviewable output
Token delta
measured delta
Evidence receipt
Run the measurement to see source size, output size, warnings, and TOON.
Source chars
0
~0 tokens
TOON chars
0
~0 tokens
Token delta
0
0%
Optimizer warnings
duplicate rules, vague rules, long sections, split candidatesWarnings appear after measurement.
TOON output
How it works
Profile first. Encode second. Measure always.
CheapAgent uses the doc2toon browser package as its engine. The UI is the hosted app; doc2toon remains the conversion boundary.
Detect structure
The engine profiles definitions, requirements, rules, tables, and mixed agent-context documents before encoding.
Convert to TOON
Record-like content can remove repeated field names. Prose may get larger, and the receipt says so when it does.
Surface warnings
Duplicate, vague, long, and split-candidate warnings are advisory signals. CheapAgent does not silently rewrite your source.
Claims boundary
Measure before you claim savings.
CheapAgent is an early alpha for browser-side agent-context measurement. Savings are measured per input and are not guaranteed.
- It does process source text locally in the browser.
- It does not call a hosted LLM API from this page.
- It does not store document bodies server-side by default.
- It is not a compliance, legal-assurance, or production-reliability certification system.