How databricks-solutions/lakebase-for-ai-developers works
This is a manufacturing supply-chain planner copilot: a planner submits a question, the graph routes it to parallel knowledge, analytics, and operational gathering, then hydrates memory, proposes a plan, and pauses for approval when needed before commit. The React app provides chat, review, and backend exploration; Lakebase backs resumable state and memory, while Genie and retrieval services supply evidence. Evaluation and data preparation are separate product-support branches. Some subsystem links are documented in the README; unsampled implementations are represented only at paths supported by the tree or source index.
Main entry points: planner.py Plan and gate · App.tsx React app · gather_nodes.py Gather agents · agent.py Agent adapter
Components 28
Planner experience
- React app
frontend/src/App.tsx - Chat panel
frontend/src/components/ChatPanel.tsx - Approval review
frontend/src/components/ReviewPanel.tsx - Backend explorer
frontend/src/components/ExplorerDrawer.tsx - Agent API
agent_server/webapp.py
Agent workflow
- Agent adapter
agent_server/agent.py - Graph assembly
agent_server/graph/build_graph.py - Supervisor routing
agent_server/graph/supervisor.py - Plan and gate
agent_server/graph/planner.py - Human approval
agent_server/graph/planner.py - Shared contracts
agent_server/contracts.py - Runtime settings
agent_server/config.py
Evidence retrieval
- Gather agents
agent_server/graph/gather_nodes.py - Knowledge retrieval
agent_server/tools/knowledge_tool.py - Genie analytics
agent_server/tools/genie_tool.py - Operational retrieval
agent_server/operational_db.py
State and memory
- Memory hydration
agent_server/graph/memory_nodes.py - Memory policy
agent_server/memory.py - Lakebase state
agent_server/lakebase.py
Data and evaluation
- Genie setup
data/genie/genie_config.py - Knowledge ingestion
data/knowledge - Operational data setup
data/operational - Agent evaluation
agent_server/evaluation/runners.py - Evaluation reports
agent_server/evaluation/report.py
Other components
- Planner
- Databricks Genie
- Mosaic Vector Search
- MLflow tracing
Connections 32
- Planner → React app: asks
README.md - React app → Chat panel: renders
frontend/src/App.tsx - Chat panel → Agent API: sends chat
frontend/src/components/ChatPanel.tsx - React app → Agent API: resumes approval
frontend/src/App.tsx - React app → Approval review: presents review
README.md - React app → Backend explorer: opens explorer
frontend/src/App.tsx - Agent API → Agent adapter: serves agent
agent_server/webapp.py - Agent adapter → Graph assembly: runs graph
agent_server/agent.py - Graph assembly → Supervisor routing: starts routing
agent_server/graph/build_graph.py - Supervisor routing → Gather agents: dispatches
agent_server/graph/build_graph.py - Gather agents → Knowledge retrieval: retrieves knowledge
- Gather agents → Genie analytics: queries analytics
- Gather agents → Operational retrieval: queries operations
- Knowledge retrieval → Mosaic Vector Search: searches index
README.md - Genie analytics → Databricks Genie: queries
README.md - Operational retrieval → Lakebase state: hybrid queries
README.md - Gather agents → Memory hydration: fan in
agent_server/graph/build_graph.py - Memory hydration → Plan and gate: supplies context
agent_server/graph/build_graph.py - Memory hydration → Memory policy: uses memory
- Memory policy → Lakebase state: stores memories
README.md - Graph assembly → Plan and gate: routes workflow
agent_server/graph/build_graph.py - Plan and gate → Human approval: interrupts for review
README.md - Human approval → Plan and gate: resumes decision
README.md - Plan and gate → Lakebase state: checkpoints and commits
README.md - Agent adapter → MLflow tracing: traces run
README.md - Plan and gate → Shared contracts: uses shared shapes
agent_server/graph/planner.py - Agent adapter → Runtime settings: reads settings
agent_server/agent.py - Knowledge ingestion → Mosaic Vector Search: builds index
README.md - Genie setup → Databricks Genie: configures space
README.md - Operational data setup → Lakebase state: seeds data
README.md - Agent evaluation → Graph assembly: evaluates runs
agent_server/evaluation/runners.py - Agent evaluation → Evaluation reports: writes results