AI Platform

Lavc Systems

Local multi-agent AI orchestration platform with RAG, Kanban task management, autonomous goals, and real-time observability.

Lavc Systems operation dashboard showing local AI agents, task status, and observability signals
Operation dashboard for local AI agents, autonomous tasks, system status, and real-time observability.
Lavc Systems agent orchestration graph with multi-agent workflow nodes and execution paths
Agent orchestration graph showing how coordinator, specialist agents, memory, and tools move through a workflow.
Lavc Systems control metrics panel for local LLM runtime, memory, scheduler, and execution health
Control metrics make local LLM runtime, scheduler health, memory usage, and execution signals visible.

Project Read More

Lavc Systems

Backend Frontend LLM Orchestration RAG

Lavc Systems is a local desktop/web platform for orchestrating AI agents, managing corporate automation, storing RAG knowledge, and tracking operational work through a Kanban interface. The system runs a FastAPI backend, a React/Vite frontend, and a set of local agents connected to Ollama — all on your own machine, with no external AI service required.

BackendFastAPI, Python 3.14, Uvicorn
FrontendReact 18, TypeScript, Tailwind CSS, Vite
StateZustand
LLM RuntimeOllama · default qwen2.5-coder:7b
OrchestrationLangGraph · multi-agent queue
RAG & MemoryChromaDB · HuggingFace embeddings · HybridMemory
DatabaseSQLite, SQLAlchemy
Real-timeWebSocket
SchedulerAPScheduler + scheduler.db
SecurityUsers, companies, audit log, local vault, scopes

System Architecture

The platform separates concerns across four execution layers: the user-facing React frontend, the FastAPI backend with modular routers, the multi-agent orchestration layer, and the data layer covering SQLite, ChromaDB, and local file storage.

React Frontend (Vite) → FastAPI Routers → Task Queue
Coordinator Agent → LangGraph Pipeline → Specialist Agents
HybridMemory + ChromaDB → Ollama (local LLM) → WebSocket Events → UI

Operational Flow

User creates task → Kanban + queue → Coordinator picks pipeline
Agents use memory, RAG & tools → Logs + artefacts on task detail
Operational trace → Auditable: intent / context / action / next step

Interface Modules

ScreenPurpose
OperationKanban, Scheduler and Objectives: tasks, recurring jobs and prioritised goals.
ControlExecutive dashboard, system health, resources, logs, traces and execution.
KnowledgeRAG documents, continuous learning, memories, skills and strategies.
AgentsAgent Graph (primary), Agent Studio and Workflow for editing and designing flows.
ToolsPlugins, APIs and tool registry available to agents.
SettingsModels, runtime parameters and system preferences.

Default Agent Roster

TypeRole
coordinatorPlans, decides pipeline, delegates.
analystAnalyses data, documents and context.
programmerImplements and adjusts code.
executorRuns commands, scripts and validations.
documenterProduces documentation and text artefacts.
reviewerReviews quality, consistency and final result.
specialist:*Domain specialists created in Agent Studio.

RAG and Knowledge

The Knowledge screen groups all knowledge management in one place:

  • RAG Documents: upload, text editing, metadata, deletion and reindexing.
  • Continuous Learning: experiences, skills, strategies and memories.
  • Hybrid Memory: SQLite for structured records and ChromaDB for semantic search.
  • Task artefacts: generated results and files can feed back into the internal knowledge base.

Skyler Inside Lavc Systems

Skyler is the floating AI assistant embedded in the system. In Lavc Systems, Skyler operates at a significantly more advanced level than the public portfolio version: she can query system health, read task data, create and execute tasks, read uploaded files, save API credentials to the local vault, and display an auditable operational trace for each response. The system does not record raw model reasoning — what surfaces is a structured summary: detected intent, context consulted, tool or action used, and next step.

The public portfolio Skyler at /skyler-assistant/ is a lighter, cloud-deployed version intended for external visitors. Lavc Systems Skyler runs locally, has access to internal data, and participates directly in the task orchestration loop.

Main API Endpoints

MethodRouteDescription
GET/POST/api/tasksList or create tasks.
POST/api/tasks/{id}/executeEnqueue execution.
GET/api/tasks/{id}/logsLogs and operational traces.
GET/POST/api/tasks/{id}/messagesPersistent task chat.
GET/api/agentsRegistered agents.
GET/POST/api/documentsRAG documents.
POST/api/documents/uploadUpload and index.
GET/POST/api/scheduler/jobsScheduled jobs.
GET/POST/api/objectives/goalsAutonomous goals.
POST/api/assistant/chatSkyler endpoint.
WS/wsReal-time events.

Project Structure

vale-develop-analytics-2026/
  backend/
    agents/          # LangGraph pipelines and multi-agent orchestration
    app/             # global subsystem state
    auth/            # users, companies and authentication
    knowledge/       # graph, semantic index and solution library
    learning/        # experiences, strategies, skills and prompt optimizer
    memory/          # HybridMemory
    objectives/      # GoalEngine, priority and autonomous scheduler
    observability/   # traces, metrics and execution
    routers/         # modular endpoints
    scheduler/       # scheduled jobs
    tools/           # registry, tools and Graphify
    main.py          # FastAPI entry point
    ws.py            # WebSocket
  frontend/
    src/
      components/    # shared UI, graph, workflow, chat
      hooks/         # WebSocket and Agent Graph runtime
      pages/         # app screens
      store/         # Zustand stores
      types/         # TypeScript types
  plugins/           # local plugins
  data/              # databases, logs, uploads, vectors and local artefacts

Running Locally

Windows:

start.bat

Linux / macOS:

./start.sh
  • Frontend: http://localhost:5173
  • Backend: http://localhost:8000
Open Skyler Assistant (Portfolio Version)