I-Lang Compress

An AI-native prompt compression protocol created by a Chinese developer.

Compress natural language prompts into dense structured instructions that any AI understands natively. 40-65% token savings, zero training needed.

> Safety note: I-Lang output is text notation, not executable code. The instruction set includes action verbs (READ, WRIT, DEL, COPY, SYNC) and resource references (@GH, @DRIVE, @LOCAL) that describe operations in compressed form but do not execute by themselves. If you pass compressed output to an agent or tool that interprets these as commands, real actions may be triggered. Always review before feeding to execution environments.

Why I-Lang

Token is money. Every prompt you send to GPT/Claude/Gemini, you pay by token. I-Lang compresses your instructions into a fraction of the original size — AI reads it just as well, you pay less.

How to compress

When the user asks to compress a prompt, convert it to I-Lang syntax following these rules.

Syntax

Single operation: [VERB:@ENTITY|mod1=val1,mod2=val2]

Pipe chain: [VERB1:@SRC]=>[VERB2]=>[VERB3:@DST]

Each step receives previous output as @PREV.

Available Verbs (62)

Data I/O: READ, WRIT, DEL, LIST, COPY, MOVE, STRM, CACH, SYNC, Π

Transform: Σ, Δ, φ, ∇, DEDU, ∂, CHNK, FLAT, NEST, λ, REDU, PIVT, TRNS, ENCD, DECD, ξ, ζ, EXPN, θ, FMT

Analysis: ψ, CLST, SCOR, BNCH, AUDT, VALD, CNT, μ, TRND, CORR, FRCS, ANOM

Generation: CREA, DRFT, PARA, EXTD, SHRT, STYL, TMPL, FILL

Output: Ω, DISP, EXPT, PRNT, LOG

Meta: VERS, HELP, DESC, INTR, SELF, ECHO, NOOP

Modifiers (28)

tgt, src, dst, frm, to, scp, dep, rng, whr, mch, exc, lim, off, top, bot, fmt, lng, sty, ton, len, col, row, srt, grp, typ, enc, chr, cap

Entities (14)

@R2, @COS, @GH, @DRIVE, @LOCAL, @WORKER, @CF, @SCREEN, @LOG, @NULL, @STDIN, @SRC, @DST, @PREV

Compression Guidelines

• Output the compressed I-Lang instruction first, then a brief explanation of what each step does.
• Use pipe chains for multi-step operations.
• Use Greek symbols where applicable (Σ for merge, Δ for diff, φ for filter, etc.)
• Maximize compression while preserving complete semantics.
• If input is ambiguous, ask the user for clarification.

Examples

Input: Read the config file from GitHub and format it as JSON

Output: [READ:@GH|path=config.json]=>[FMT|fmt=json]

Explanation: READ fetches from GitHub, FMT converts to JSON format.

Saved: 55%

Input: Filter all fatal errors from system logs

Output: [φ:@LOG|whr="lvl=fatal"]

Explanation: φ (filter) selects only entries matching fatal level.

Saved: 55%

Input: Read all markdown files, merge them, summarize in 3 bullets, output

Output: [LIST:@LOCAL|mch="*.md"]=>[Π:READ]=>[Σ|len=3]=>[Ω]

Explanation: LIST finds files, Π batch-reads, Σ summarizes to 3 items, Ω outputs.

Saved: 65%

Links

• Homepage: https://ilang.ai
• Dictionary: https://github.com/ilang-ai/ilang-dict

Author

Built by ilang-ai from China. I-Lang is open source under MIT license.

I-Lang v2.0