You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Q: What is OSMP?
A: A wire protocol that compresses agentic AI instructions into a deterministic encoding any device can decode by table lookup. No inference at the decode layer. 342 opcodes across 26 namespaces. Works across any transport, from a 51-byte LoRa packet to a cloud pipeline.
Q: How is this different from MCP, A2A, or ACP?
A: Those are framework-layer protocols. OSMP is an encoding-layer protocol. It sits beneath any of them as the serialization format. Two agents on different frameworks that share the OSMP dictionary can communicate without modifying either framework.
Do I need to change my agent framework to use it?
No. The MCP server (pip install osmp-mcp) gives your agent encode/decode as tools. One config line. Your framework stays the same.
Q: How do I test it?
A: python3 sdk/python/src/osmp.py runs the benchmark against the canonical test vectors. You'll see per-instruction byte counts and the mean reduction. The MCP server has a benchmark tool built in. The 29-vector framework comparison runs from benchmarks/sal-vs-json/.
Q: What SDKs are available?
A: Python (reference), TypeScript (npm: osmp-protocol), Go. All three are independently verified against the same canonical test suite.
Q: Where are the compression numbers from?
A: 29 real-world vectors from MCP, OpenAI, A2A, CrewAI, and AutoGen. Full methodology, .proto schemas, and reproduction scripts in docs/SAL-efficiency-analysis.md. Run them yourself.
Q: Is this production ready?
A: The protocol works. Three conformant SDKs, MCP server on PyPI, encoding and decoding verified against a canonical test suite. There are zero production deployments so far and no independent audit. If you deploy it, you may be the first. We want to hear what you learn.
Q: What does "patent pending" mean for me?
A: Apache 2.0 with express patent grant. If you implement this spec, you have a license. The patent applies to non-conformant implementations that use equivalent methods. Build on the protocol and you're covered. Full details in PATENT-NOTICE.md.
Q: What's not built yet?
A: C++ firmware-level encoder/decoder for ESP32/nRF52 sovereign nodes. That's the highest-priority contribution target.
reacted with thumbs up emoji reacted with thumbs down emoji reacted with laugh emoji reacted with hooray emoji reacted with confused emoji reacted with heart emoji reacted with rocket emoji reacted with eyes emoji
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
-
Q: What is OSMP?
A: A wire protocol that compresses agentic AI instructions into a deterministic encoding any device can decode by table lookup. No inference at the decode layer. 342 opcodes across 26 namespaces. Works across any transport, from a 51-byte LoRa packet to a cloud pipeline.
Q: How is this different from MCP, A2A, or ACP?
A: Those are framework-layer protocols. OSMP is an encoding-layer protocol. It sits beneath any of them as the serialization format. Two agents on different frameworks that share the OSMP dictionary can communicate without modifying either framework.
Do I need to change my agent framework to use it?
No. The MCP server (pip install osmp-mcp) gives your agent encode/decode as tools. One config line. Your framework stays the same.
Q: How do I test it?
A: python3 sdk/python/src/osmp.py runs the benchmark against the canonical test vectors. You'll see per-instruction byte counts and the mean reduction. The MCP server has a benchmark tool built in. The 29-vector framework comparison runs from benchmarks/sal-vs-json/.
Q: What SDKs are available?
A: Python (reference), TypeScript (npm: osmp-protocol), Go. All three are independently verified against the same canonical test suite.
Q: Where are the compression numbers from?
A: 29 real-world vectors from MCP, OpenAI, A2A, CrewAI, and AutoGen. Full methodology, .proto schemas, and reproduction scripts in docs/SAL-efficiency-analysis.md. Run them yourself.
Q: Is this production ready?
A: The protocol works. Three conformant SDKs, MCP server on PyPI, encoding and decoding verified against a canonical test suite. There are zero production deployments so far and no independent audit. If you deploy it, you may be the first. We want to hear what you learn.
Q: What does "patent pending" mean for me?
A: Apache 2.0 with express patent grant. If you implement this spec, you have a license. The patent applies to non-conformant implementations that use equivalent methods. Build on the protocol and you're covered. Full details in PATENT-NOTICE.md.
Q: What's not built yet?
A: C++ firmware-level encoder/decoder for ESP32/nRF52 sovereign nodes. That's the highest-priority contribution target.
Beta Was this translation helpful? Give feedback.
All reactions