SOMA is a decentralized AI solutions subnet operating as Subnet 114 (SN114) on the Bittensor network. Built by Dendrite Holdings, SOMA delivers practical AI capabilities through MCP (Model Context Protocol) infrastructure, a standardized way for AI agents to connect to external tools, data sources, and services. Just as USB-C standardizes how devices connect to peripherals, MCP standardizes how agents connect to capabilities. SOMA's MCP servers act as an integration layer that aggregates inputs, manages state, and coordinates execution across decentralized systems, turning fragmented agent outputs into coherent, reliable behavior. The subnet operates on a competition-based model. For each problem that can meaningfully improve agent performance and be solved using MCP, SOMA organizes a competition. Independent miners compete to build the best algorithms, and the winning solution gets deployed as a production MCP server on the SOMA platform. This creates a continuously evolving marketplace of MCP servers that can be used by AI agents across the Bittensor ecosystem, as well as by open-source tools like OpenClaw. The result is practical, plug-and-play access to standardized services that improve agent outputs, reduce integration complexity, and accelerate iteration for builders and miners across subnets. The first competition series focuses on Context Compression, one of the fundamental bottlenecks in modern agent architectures. As AI moves from simple chatbots to complex multi-step agents, token costs are becoming a defining constraint. A single agentic interaction can consume tens of thousands of input tokens as conversation history and retrieved context accumulate with every reasoning step, driving up costs even on the cheapest models. Research has also shown that LLM performance can drop significantly as context length increases, meaning longer prompts can produce worse outputs while costing more.

Not Available