HiveFlow Modules¶
HiveFlow ships as three cooperating packages in one repository. Each has its own README with install steps and API entry points.
Overview¶
flowchart TB
subgraph Studio["Studio — packages/studio"]
UI[React UI]
API[FastAPI backend]
end
subgraph Agent["Agent — hiveflow-agent"]
HM[HiveMindApp]
CO[CognitiveOrchestrator]
end
subgraph Core["Core — hiveflow"]
SCH[Scheduler]
BB[Blackboard]
ORCH[Orchestrator / HITL]
end
UI --> API
API --> HM
API --> ORCH
HM --> CO
CO --> SCH
CO --> BB
ORCH --> SCH
ORCH --> BB
Core — orchestration kernel¶
PyPI: hiveflow · Path: packages/core
Embeddable Python engine: event bus, skill-based scheduler, audited blackboard, static/dynamic DAG execution, HITL gates, checkpoints, guards, RAG, and MCP — with no UI or LLM requirement.
Use when: you integrate orchestration into your own service and control agents/ECM directly.
→ Full Core README on GitHub · API · Architecture
Agent — cognitive runtime¶
PyPI: hiveflow-agent · Path: packages/agent · Requires: hiveflow
Adds HiveMindApp: natural language → TaskGraph plan → Skill-bound ReAct workers. Supports run_query, plan_only, and execute_plan (including plans imported from Studio or LangGraph).
Use when: users describe tasks in NL, or you want automatic planning/replanning on top of Core.
→ Full Agent README on GitHub · Studio Agent cookbook
Studio — visual operations platform¶
Path: packages/studio · Not on PyPI · Stack: FastAPI + React
Self-hosted UI: Orchestrator, Chatflow, Approvals (HITL), Analytics, Tracer, Replay. Backend switches between Core DAG and Agent runtime (HIVEFLOW_RUNTIME=agent).
Use when: operators, reviewers, or demos need a browser-based control plane.
→ Full Studio README on GitHub · Agent ops
Which module do I need?¶
| Goal | Modules |
|---|---|
| Python library in my app | Core only |
| NL agent with planning | Core + Agent |
| Team UI + HITL + analytics | Core + Agent + Studio (typical full stack) |
| Fixed DAG, no LLM | Core (+ Studio optional for canvas) |