| license: mit | |
| language: en | |
| library_name: transformers | |
| tags: | |
| - nullstate | |
| - agent-intelligence | |
| - payment-infrastructure | |
| - x402 | |
| - ap2 | |
| - agent-economy | |
| # NullState Intelligence | |
| NullState's AI model for agent task intelligence, scoring, and settlement orchestration. | |
| ## Overview | |
| NullState uses a dual-model architecture: | |
| - **Phi-3-mini-4k-instruct** (local, Microsoft) - 20% of queries | |
| - **Google Gemini 2.0 Flash** (cloud) - 80% of queries | |
| - Keyword fallback - ~2% when both unavailable | |
| ## Usage | |
| ```python | |
| from nullstate_intelligence import NullStateModel | |
| model = NullStateModel() | |
| result = model.score_task("Calculate 15% APR on 1000 USDC for 30 days") | |
| ``` | |
| ## Training Data | |
| Model training data is collected from live gateway telemetry at `greensol.me/nullstate`. Each interaction is labeled with: | |
| - Task type classification | |
| - Response quality score (1-5) | |
| - Revenue stream attribution | |
| - Latency metrics | |
| ## Protocol Support | |
| The model understands and generates: | |
| - x402 payment challenges | |
| - AP2 mandate structures | |
| - MCP JSON-RPC tool calls | |
| - KYA challenge/response pairs | |
| ## Deployment | |
| Deployed as part of the NullState gateway stack. Dual-model with graceful degradation. | |
| [Full documentation →](https://greensol.me/nullstate/docs/architecture) | |