Compress Prose to AXL v3

Deterministic compression into structured packets. No LLM needed. Fewer tokens, lower cost.

Free, forever. Apache 2.0, community-stewarded by AXL Protocol. Sponsors keep the lights on and unlock unlimited usage - nobody pays to use the protocol.

AXL Compress hero illustration
English Input
0 chars
AXL v3 Output
0 chars | 0 words
Tokens Saved -- Cost (Sonnet) -- Packets --
SPONSOR Sign in for API keys

Recent Compressions

No compressions yet

How It Works

1

Paste Text

Drop in any English prose - docs, articles, instructions, transcripts.

2

Compress

Deterministic NLP pipeline parses your text into structured AXL v3 packets. No LLM.

3

Save Tokens

Fewer tokens = lower API costs. Deterministic, reversible, round-trip proven.

Why Not Just Strip Tokens?

TOKEN STRIPPING
patient elevated CA-125 285 U/mL 8.1x baseline complex ovarian mass CT endometriosis history malignancy 60% MRI biopsy 2 weeks

No structure. No types. No confidence scores.

No operation codes. No evidence chains.

Cannot decompress back to prose.

Ratio
StructureNone
Round-tripNo
TypesNo
MetricsNo
AXL v3 COMPRESSION
ID:ONC-01|OBS.99|@patient|^CA125:285+^baseline:8.1x|NOW ID:RAD-01|OBS.95|@patient|^mass:complex|CT|NOW ID:ONC-01|INF.80|@diagnosis|<-^CA125+^imaging|~malignancy_60%|1W ID:PATH-01|CON.70|@dx|RE:ONC-01|<-^endo_hx|1W ID:TEAM|MRG.85|@tx|RE:all|^MRI+biopsy|1W

Typed subjects. Confidence scores. Evidence chains.

7 operations. 6 tag types. Machine-parseable.

Decompresses back to full prose.

Ratio1.96x
StructureCompositional grammar
Round-tripYes
Types$ @ # ! ~ ^
MetricsFull (ratio, packets, saved)

Same source text. One strips words. The other compresses meaning.

Full AXL Chat Pipeline

Sign in for the 3-step chat experience. Your AI thinks in compressed AXL. You chat in English. Every message runs through the compress-reason-decompress pipeline, saving tokens on every exchange.

Sign In to Access See 7 Examples
Architecture and Internals

Rosetta v3 Kernel

The Rosetta is a 75-line specification that defines the AXL v3 packet grammar. It is fetched once from axlprotocol.org/v3 and cached in browser memory. When you press Compress, it becomes the system prompt for the LLM call.

Packet Format

Every AXL v3 packet follows this structure:

ID:AGENT | OP | subject + predicate + object [meta]

Operations: OBS (observation), INF (inference), CON (contradiction), MRG (merge/synthesis), SEK (knowledge gap), YLD (yield/update), PRD (prediction).

Subject Tags

$ financial @ entity # metric ! event ~ state ^ value

Evidence and References

<- source (evidence chain) RE: packet-ref

API Key Security

Your API key is stored only in your browser tab's memory. It is sent directly to the LLM provider (Anthropic or OpenAI) via HTTPS. Our server never sees, stores, or proxies your key for this tool. The registered Chat pipeline stores keys server-side for the 3-step compress/reason/decompress flow.

Decompression

To decompress AXL packets back to English: parse each packet by operation type, extract the tagged subjects and predicates, reconstruct claims in natural language, then merge into coherent prose following the MRG packets as structural guides.

Visual Guide