Octid

OSMP v3.0.0 released. Five things this protocol carries: Reach, Intent, Proof, Computation, Consequence. Each moved this release. Reach: Four SDKs at full feature parity: Python, TypeScript, Go, Rust. Rust at 0.5.2 with the full ADP / EML / MDR / FNP / SEC stack — same wire bytes as the other three on every commit. The protocol runs wherever your stack runs. Intent: SAL deterministic decode. 35 bytes on the wire for a typical instruction. Decoded by dictionary lookup, not inference. The instruction is the intent. Proof: SHA-256 ASD fingerprint (9ecc507e2c24c4a7) verified identical across all four SDKs in CI on every commit. Any drift between Python, TypeScript, Go, or Rust fails the merge. SEC envelope adds strict-monotonic inbound replay protection — replayed frames refused before signature verification. Computation: Compute-by-wire. 89-function EML registry public across all four SDKs. Pre-built eml(x, y) = exp(x) − ln(y) chain templates spanning compound arithmetic, neural-network activations, linear algebra, trigonometry, and 19 scientific compounds. A full sin(x) or sqrt(x) approximation fits in fewer than 100 bytes on the wire. The receiver decodes the tree and evaluates — no math library dependency. The math doesn't live anywhere as a library. It's transmitted. The byte-identical guarantee is structural, not engineered. Correctly-rounded math has a unique answer; the universal-operator architecture plus IEEE-754 determinism force convergence regardless of coefficient choice. Two independent derivations — INRIA CRLibm coefficients vs a clean-room Remez (1934) at 150-digit precision in pure Python — produce SHA-256-identical fingerprints. Cross-implementation verification by construction. Fast mode is fdlibm-derived, 1 ULP — byte-identical across all four SDKs. Precision mode is the same protocol with extended internal precision, correctly-rounded — for regulated industries (medical IEC 62304, aerospace DO-178C, nuclear IEC 61513), audit-grade finance, and cryptographic protocol-frame inputs. Commercial license at ack@octid.io. Universal operator from Odrzywolek's eml paper (arXiv:2603.21852, CC BY-4.0). OSMP builds the wire-transmission layer. Consequence: Cross-SDK semantic-safety primitive. ADP REPLACE deltas require FLAG_CRITICAL for guaranteed retransmission per spec §6/§7. v3.0.0 ships validate_received_delta canonical across all four SDKs — receivers invoking it reject under-flagged REPLACE deltas with a canonical DeltaValidationError. Spec-mandated semantics are no longer SDK-specific. Versions: osmp 2.8.0 (PyPI) osmp-mcp 1.3.4 (PyPI) osmp-protocol 2.8.0 (npm) osmp 0.5.2 (crates.io) GitHub release: https://lnkd.in/eu3kUm3X EML universal operator paper (Odrzywolek, CC BY-4.0): https://lnkd.in/e559iUH7 The open encoding standard for agentic instruction — from intent to outcome. Patent pending. Apache 2.0 with express patent grant.

OSMP v2.0.0 released. One month since filing. Here’s what shipped… Breaking: v15 dictionary. 24 opcode renames validated against frontier model recognition convergence. 352 opcodes, 26 namespaces. Meta-verb opcodes removed — opcodes name what is being operated on, not the act of operating on it. New: Deterministic NL-to-SAL composition. SALComposer takes natural language in and produces validated SAL out. No inference in the pipeline. 358 phrase triggers. 95.7% opcode coverage on the full dictionary. The deterministic tool scores 91.7% composition fidelity — outperforming frontier LLMs given the full composition doctrine (77-81%). Ported to Python, TypeScript, Go. Three new MCP tools bring the server to 17 total. New: SAIL basis-gated capability negotiation. Two nodes exchange 8-byte dictionary fingerprints before binary mode. Matching bases get SAIL. Mismatched bases degrade to SAL. No silent misdecode. The intern table is a pure function of the basis. Implemented across all three SDKs. New: Registered macro architecture. 16 Meshtastic macros ship as pre-validated SAL chain templates. Slot-fill invocation, compact wire format, consequence class inheritance. A multi-step workflow becomes one lookup. New: Behavioral vs actuator command layers (ADR-005). R:ACC/R:THR, R:DECEL/R:BRK — planning-layer commands and control-layer commands are now distinct at the wire format. Every production autonomous vehicle stack maintains this separation. Now the protocol does too. Spec updated: Principles 3 and 4 (no meta-verbs, namespace-mnemonic alignment), composition Rule 10, sections 3.6 and 9.1–9.8 revised. Cross-SDK wire compatibility: 86/86 vectors, all three SDKs. 624 tests passing. ASD drift check clean. They compress text. We encode instructions. Patent pending. Apache 2.0 with express patent grant. GitHub: https://lnkd.in/gPtcBpR5 Patent pending. Apache 2.0 with express patent grant.

OSMP v1.0.0 tagged and released. Open wire protocol for agentic AI instruction exchange. 342 opcodes, 26 namespaces, three conformant SDKs, MCP server on PyPI. 76% fewer tokens than JSON-RPC. 86.8% fewer bytes than JSON. 70.5% smaller than compiled Protocol Buffers. Measured, published, reproducible. Confirmed over LoRa mesh radio today. Four instruction types encoded, transmitted, and decoded by table lookup across two Meshtastic nodes. No inference at the decode layer. No cloud. No firmware changes. Patent pending. Apache 2.0 with express patent grant. GitHub: https://lnkd.in/g8DNZbGR

OSMP (Octid Semantic Mesh Protocol) is an open wire protocol for agentic AI instruction exchange. It encodes inter-agent instructions 86.8% smaller than JSON and cuts LLM token consumption by 76%. Decode is deterministic table lookup. No inference at the receiving node. Any device, any channel. From a 51-byte LoRa radio packet to cloud-scale agent orchestration. The whitepaper covers the full structural efficiency analysis across five production frameworks, including a four-way comparison against JSON, MessagePack, and Protocol Buffers. Whitepaper: https://lnkd.in/g-mT3ZE9 Repo: https://lnkd.in/gKazQ3me Site: https://octid.io