How ivancordonm/agentic-platform-starter-kit works
This is a configuration-first platform for building and running agent applications. The main workflow is: a caller submits input to the API, the API selects a versioned project runtime, and either an individual agent or a configured graph/orchestrator workflow executes; run results and events are retained for inspection. Project loading validates configuration and resolves models, prompts, schemas, and tools before execution. The Streamlit UI uses the API for project controls, execution, and traces. The README documents additional workflow modes, authentication, reload, and optional SQLite storage; unsampled details are represented only where documented.
Main entry points: routes.py API routes · project_runtime.py Project runtime · agent_executor.py Agent executor · runtime_manager.py Runtime manager
Components 22
API and UI
- Developer UI
app/ui/streamlit_app.py - HTTP client
app/ui/client.py - API routes
app/api/routes.py - HTTP contracts
app/api/models.py
Runtime and configuration
- Runtime manager
app/engine/runtime_manager.py - Project runtime
app/engine/project_runtime.py - Config loading
app/engine/config.py - Config definitions
app/engine/definitions.py - Agent registry
app/engine/agent_registry.py - Model registry
app/engine/model_registry.py - Prompt loading
app/engine/prompt_loader.py - Schema loading
app/engine/schema_loader.py - Tool registry
app/engine/tool_registry.py - Project definitions
project
Agent execution
- Agent executor
app/engine/agent_executor.py - Agent factory
app/engine/agent_factory.py - Workflow executor
app/engine/workflow_executor.py
Run records
- Run repository
app/engine/run_store.py - SQLite run store
app/engine/sqlite_run_store.py
Other components
- API caller
- OpenAI model service
- MCP tool service
Connections 26
- API caller → API routes: submits requests
README.md - Developer UI → HTTP client: uses
app/ui/streamlit_app.py - HTTP client → API routes: calls API
README.md - API routes → HTTP contracts: validates requests
app/api/routes.py - API routes → Runtime manager: selects snapshot
app/api/routes.py - API routes → Agent executor: runs agent
app/api/routes.py - API routes → Workflow executor: runs workflow
app/api/routes.py - Runtime manager → Project runtime: builds runtime
app/engine/runtime_manager.py - Runtime manager → Agent executor: builds executor
app/engine/runtime_manager.py - Runtime manager → Workflow executor: builds executor
app/engine/runtime_manager.py - Project runtime → Config loading: loads configuration
app/engine/project_runtime.py - Project runtime → Agent registry: initializes registry
app/engine/project_runtime.py - Project runtime → Model registry: initializes registry
app/engine/project_runtime.py - Project runtime → Prompt loading: initializes loader
app/engine/project_runtime.py - Project runtime → Schema loading: initializes loader
app/engine/project_runtime.py - Project runtime → Tool registry: initializes registry
app/engine/project_runtime.py - Config loading → Config definitions: validates against
app/engine/config.py - Config loading → Project definitions: reads project config
README.md - Agent executor → Agent factory: builds agent
app/engine/agent_executor.py - Agent executor → Tool registry: opens MCP tools
app/engine/agent_executor.py - Agent executor → OpenAI model service: invokes model
README.md - Tool registry → MCP tool service: connects to MCP
README.md - Workflow executor → Agent executor: runs workflow agents
app/engine/workflow_executor.py - API routes → Run repository: records run events
app/api/routes.py - API routes → SQLite run store: selects when configured
app/api/routes.py - API routes → HTTP client: serves API responses
README.md