Migrating from LangGraph¶
Concept mapping between LangGraph and HiveFlow. HiveFlow targets teams that need HITL, Studio UI, and secure shared state out of the box.
Concept map¶
| LangGraph | HiveFlow | Notes |
|---|---|---|
StateGraph |
DAGOrchestrator / CognitiveOrchestrator |
Static vs runtime-generated plans |
| Graph nodes | Agents + task_handler |
Skills-based scheduling |
| Shared state | SecureBlackboard |
Encrypted + audited keys |
interrupt() / human node |
HITLManager + HITLAction |
Native gates + Studio Approvals |
| Checkpointer | CheckpointManager |
Blackboard snapshots |
| Tool calling | MCP + ReActWorker |
MCP-first plugin model |
| LangSmith traces | Studio Tracer / Replay | trace_id ↔ intent_id |
Typical migration paths¶
1. Fixed DAG → Core orchestrator¶
Replace LangGraph edges with a HiveFlow task graph or Studio Orchestrator canvas, then execute via Core scheduler or /api/workflows/execute.
2. Dynamic agent → Agent runtime¶
Replace create_react_agent loops with:
from app import HiveMindApp, HiveMindConfig # hiveflow-agent
result = await app.run_query("Summarize the quarterly report")
Studio equivalent: POST /api/agent/query.
3. Human approval¶
LangGraph interrupt_before → HiveFlow:
gate = await hitl.create_gate(
workflow_id="wf-1",
node_id="publish",
action=HITLAction.APPROVAL,
prompt="Approve before publish",
context={"draft": "..."},
)
See HITL Cookbook.
What is not 1:1 yet¶
- LangGraph adapter — PoC available: LangGraph integration (
taskgraph_to_langgraph/ round-trip). Full runtime bridge planned v0.3. - LangChain Tool wrappers — use MCP or custom
ReActTooltoday - LangSmith hosted eval — use HiveFlow
Evaluator+ Studio Analytics
Next steps¶
- LangGraph + HiveFlow Sidecar — recommended integration pattern
- Getting Started
- Studio Agent Mode