Feature Catalog¶
All major features of HART OS (Hevolve Hive Agentic Runtime) -crowdsourced compute infrastructure for autonomous Hive AI Training. Each entry links to a dedicated detail page.
| Feature | Description | Details |
|---|---|---|
| Agent Engine | Unified goal engine where GoalManager creates goals, AgentDaemon processes them, and SpeculativeDispatcher runs budget-gated LLM calls. | agent-engine.md |
| Recipe Pattern | Learn task execution once (CREATE mode), then replay efficiently (REUSE mode) without repeated LLM calls -- 90% faster. | recipe-pattern.md |
| Revenue Model | 90/9/1 split: 90% to users (contribution-proportional), 9% to infrastructure, 1% to central. | revenue-model.md |
| Compute Policies | Three policies (local_only, local_preferred, any) that control which models a node may use and how metered costs are handled. | compute-policies.md |
| Budget Gating | Pre-dispatch cost control combining per-goal Spark budgets, per-model token cost estimates, and platform-level affordability checks. | budget-gating.md |
| Contribution Scoring | Weighted scoring system that quantifies each node's contributions (uptime, GPU hours, content, API costs absorbed) for revenue distribution. | contribution-scoring.md |
| Metered API Recovery | Tracks and compensates nodes that spend real money on paid APIs while serving hive or idle tasks for other users. | metered-api-recovery.md |
| Federation & Gossip | Decentralized peer discovery via UDP gossip, hierarchical sync (local to regional to central), and Ed25519-signed messages. | federation.md |
| Social Platform (82 endpoints) | Full social platform with communities, posts, comments, votes, karma, encounters, referrals, gamification, and ad revenue sharing. | social-platform.md |
| Thought Experiments | Structured thought experiments with peer voting and constitutional governance for high-stakes decisions like live trading. | thought-experiments.md |
| Vision / VLM | Vision sidecar supporting MiniCPM (GPU) and MobileVLM ONNX (CPU) for visual understanding and embodied AI learning. | vision.md |
| Coding Agent (idle compute) | Dispatches coding tasks to the CREATE/REUSE pipeline during node idle time, with three pluggable tool backends. | coding-agent.md |
| Trading Agents (paper then live) | Paper trading with PaperPortfolio and PaperTrade models; live trading requires constitutional vote approval. | trading-agents.md |
| Channel Adapters (30+) | Over 30 channel adapters (Discord, Telegram, Slack, Matrix, WhatsApp, Email, SMS, and more) with a unified send/receive interface. | channels.md |