How f8fvwgzc/nexc-engine works
nexc-engine is a self-hosted visual LLM graph-engineering platform. A user sketches a graph in the browser, requests streamed LLM plan proposals, reviews and applies them, then starts a DAG run whose node outputs and artifacts can be collected. The backend exposes the API, persists domain state, schedules execution, and routes work to LLM providers, the Python agent runtime, or optional Symphony. README capabilities such as memory, collaboration, and security are retained as documented branches; links for unsampled implementations are dashed where wiring is not shown.
Main entry points: scheduler.rs DAG scheduler · router.rs HTTP API · app.py Agent runtime · service.rs LLM service
Components 20
Browser experience
- Web application
frontend/src/app/App.tsx - Graph editor
frontend/src/features/graph/canvas/canvas-engine.ts - Graph API client
frontend/src/features/graph/api.ts
Rust platform
- HTTP API
backend/src/app/router.rs - Auth and security
backend/src/http/handlers/auth.rs - Plan refinement
backend/src/engine/planner.rs - DAG scheduler
backend/src/engine/scheduler.rs - Agent orchestration
backend/src/orchestrator/mod.rs - LLM service
backend/src/llm/service.rs - LLM providers
backend/src/llm/anthropic.rs - Memory retrieval
backend/src/memory/mod.rs - Artifacts
backend/src/engine/artifacts.rs
Execution services
- Agent runtime
agent-runtime/src/nexc_runtime/api/app.py - Agent tools
agent-runtime/src/nexc_runtime/tools - Coding executor
backend/src/engine/executor/symphony.rs
Persistence and realtime
- PostgreSQL
- Data repositories
backend/src/repo/agents.rs - Realtime events
backend/src/realtime/events.rs
Other components
- User
- LLM provider
Connections 22
- User → Web application: uses
- Web application → Graph editor: renders
frontend/src/app/App.tsx - Graph editor → Graph API client: edits graph
- Graph API client → HTTP API: HTTP requests
- HTTP API → Auth and security: authenticates
- HTTP API → Plan refinement: routes plan requests
- Plan refinement → LLM service: requests completion
- HTTP API → DAG scheduler: starts runs
- DAG scheduler → Agent orchestration: assigns agents
- DAG scheduler → LLM service: executes LLM nodes
- LLM service → LLM providers: dispatches
backend/src/llm/mod.rs - LLM providers → LLM provider: calls provider
- DAG scheduler → Agent runtime: executes agent nodes
- Agent runtime → Agent tools: runs tool loop
- Agent runtime → LLM provider: requests generation
- DAG scheduler → Coding executor: dispatches coding
- DAG scheduler → Memory retrieval: retrieves context
- DAG scheduler → Artifacts: collects outputs
- HTTP API → Data repositories: serves domain data
- Data repositories → PostgreSQL: reads and writes
- HTTP API → Realtime events: streams updates
- Realtime events → PostgreSQL: fans out events