System Architecture
HART OS (Hevolve Hive Agentic Runtime) is crowdsourced compute infrastructure for autonomous Hive AI Training -an open multi-agent platform democratizing access to the best intelligence, just like the internet is open. It orchestrates distributed thought processes across crowdsourced agents, incentivizes compute providers, and creates a self-sustaining economy -so no single entity monopolizes AI. Agents are created, trained, and reused via the Recipe Pattern.
Core Pipeline
CREATE Mode: User Input --> Decompose --> Execute Actions --> Save Recipe
REUSE Mode: User Input --> Load Recipe --> Execute Steps --> Output (90% faster)
Module Map
Entry Point
| File |
Purpose |
hart_intelligence_entry.py |
Flask application, all HTTP routes, Waitress server on port 6777 |
Recipe Pipeline
| File |
Purpose |
create_recipe.py |
Agent creation, action execution, recipe generation |
reuse_recipe.py |
Recipe reuse, trained agent execution |
helper.py |
Action class, JSON utilities, tool handlers |
lifecycle_hooks.py |
ActionState machine, FlowState, ledger sync |
helper_ledger.py |
SmartLedger factory functions |
agent_ledger.py |
SmartLedger core: Task, TaskType, TaskStatus |
Agent Engine (integrations/agent_engine/)
| File |
Purpose |
goal_manager.py |
Goal creation, lifecycle management |
dispatch.py |
Goal decomposition, distributed dispatch |
parallel_dispatch.py |
Parallel task decomposition with SmartLedger |
speculative_dispatcher.py |
Fast response + background expert execution |
revenue_aggregator.py |
Revenue streams, 90/9/1 split, settlement |
budget_gate.py |
LLM cost estimation, Spark budgeting |
compute_config.py |
Compute policy resolution (env > DB > defaults) |
compute_borrowing.py |
Compute escrow service |
federated_aggregator.py |
Federated learning delta aggregation |
model_registry.py |
Model catalog, energy tracking |
| File |
Purpose |
models.py |
SQLAlchemy ORM (60+ tables), db_session() |
api.py |
82+ REST endpoints via social_bp blueprint |
services.py |
NotificationService, business logic |
peer_discovery.py |
Gossip protocol, bandwidth profiles |
hosting_reward_service.py |
Hosting rewards, contribution scoring, compute stats |
rate_limiter.py |
Token bucket rate limiting |
Security (security/)
| File |
Purpose |
master_key.py |
Ed25519 trust anchor (AI exclusion zone) |
hive_guardrails.py |
10 structurally immutable guardrail classes |
key_delegation.py |
3-tier certificate chain (central > regional > local) |
runtime_monitor.py |
Background tamper detection daemon |
node_watchdog.py |
Heartbeat protocol, frozen-thread auto-restart |
system_requirements.py |
NodeTierLevel capability detection |
| File |
Purpose |
vram_manager.py |
GPU detection, VRAM tracking, allocation |
runtime_manager.py |
Media tool lifecycle (Whisper, LTX2, MiniCPM) |
model_lifecycle.py |
Dynamic model load/unload/offload |
Other Integrations
| Directory |
Purpose |
integrations/coding_agent/ |
Coding tool orchestrator (KiloCode, Claude Code) |
integrations/vision/ |
Vision sidecar (MiniCPM + embodied AI) |
integrations/channels/ |
30+ channel adapters (Discord, Telegram, Slack, Matrix) |
integrations/ap2/ |
Agent Protocol 2 (e-commerce, payments) |
integrations/expert_agents/ |
96 specialized agents network |
integrations/internal_comm/ |
A2A communication, task delegation |
integrations/mcp/ |
Model Context Protocol servers |
integrations/google_a2a/ |
Dynamic agent registry, A2A protocol |
Database
SQLite at agent_data/hevolve_database.db with WAL mode for concurrent access. See schema.md for table details.
See Also