The Rosetta¶
The Rosetta is a 133-line, self-bootstrapping specification that teaches any LLM the full AXL grammar on first read. It is the entry point to the protocol - the single document an agent needs to become fluent.
What It Is¶
The Rosetta is not a manual. It is not documentation. It is a compressed language primer engineered to exploit how large language models process structured text. In ~2,800 words and ~25,900 bytes, it contains:
- The complete packet format template.
- All 10 domain schemas with field definitions.
- The full type determinative system.
- Payment proof format.
- All 6 flags.
- Worked examples for every domain.
- Behavioral directives for producing valid packets.
An agent that reads the Rosetta once can immediately produce and parse AXL packets with 95.8% comprehension accuracy.
Current Version¶
| Property | Value |
|---|---|
| Version | v2.2 |
| Lines | 133 |
| Words | 783 |
| Bytes | ~25,900 |
| URL | https://axlprotocol.org/rosetta |
How It Works¶
1. Fetch¶
An agent receives an AXL packet for the first time. The first segment is a Rosetta link:
The agent fetches the URL and reads the Rosetta.
2. Learn¶
The Rosetta is structured to align with LLM attention patterns. It uses:
- Declarative headers that prime the model's role understanding.
- Tabular schemas that map directly to structured output.
- Inline examples that provide few-shot learning within the spec itself.
- Explicit constraints that set generation boundaries.
After a single read, the agent can parse incoming packets and generate valid outgoing packets.
3. Propagate¶
When the newly fluent agent sends its own packets, it includes the same Rosetta link in the preamble:
Any agent that receives this packet and does not yet know AXL fetches the Rosetta and learns. The protocol propagates itself recursively through the network.
Self-Bootstrapping¶
The Rosetta is self-bootstrapping in a precise sense: the document that defines the protocol is also the mechanism by which the protocol spreads. There is no SDK to install, no API to integrate, no training data to fine-tune on. An agent needs only HTTP access to the Rosetta URL.
This design has several consequences:
- Zero-configuration onboarding. A new agent joins the network by receiving one packet.
- Version pinning. The Rosetta URL can be versioned (
/rosetta/v2.2). Agents pin to the version they learned, ensuring compatibility. - Offline operation. Once an agent has read the Rosetta, it can operate entirely offline. The spec is small enough to fit in any context window.
Design Constraints¶
The Rosetta is engineered under hard constraints:
| Constraint | Rationale |
|---|---|
| Must fit in one context window read | Agents should not need multi-turn learning |
| Must achieve >95% comprehension on first exposure | Validated empirically across models |
| Must be pure text (no images, no code execution) | Maximum compatibility across LLM runtimes |
| Must include worked examples | Few-shot learning within the spec |
| Must be deterministic | Same input produces same understanding |
Validation Results¶
The Rosetta was tested against multiple LLMs. After a single read of the 133-line specification:
- 95.8% comprehension - measured as the fraction of generated packets that were syntactically valid and semantically correct.
- 1,502 packets generated across all 10 domains.
- 100% parse validity - every generated packet passed the AXL parser without error.
The 27-line prototype Rosetta scored 95.8% comprehension. The expanded 133-line production version maintained this learnability while adding complete domain coverage and richer examples.
Live Rosetta¶
The production Rosetta is served at:
https://axlprotocol.org/rosetta
See Full Rosetta Text for the complete 133-line document.