Skip to content

Example Graphs

The demo applications register the following graphs. They demonstrate LGOS features; none is installed as a built-in model by the langgraph-openai-serve package. Each registration also supplies a short GraphConfig.description used by the demo model catalogs.

Model Demonstrates Graph feature Graph-specific dependency
deep-research Direct answers with attachments and gateway tools, or reviewed web research by parallel researcher subgraphs ending in a cited report, in the foreground or background background, file_inputs, interrupts, mcp_tools Responses model, selected gateway, web search, and PostgreSQL
background-interrupt Deterministic background report preparation, human approval, and background resumption background, interrupts Hatchet and PostgreSQL checkpointer and run coordinator
background-mock Deterministic background execution in an independently deployed worker, with no model call background Hatchet
custom-input-output-context Request, output, and typed runtime-context adapters None None
citation-events Structured OpenAI URL citations with portable Markdown content None None
coding-agent Coding agent with shell execution, file editing, and a persistent workspace None Coding-agent service (Codex) and Responses model
file-input Central Files API IDs resolved into OpenAI Responses file inputs file_inputs Central Files API and upstream Responses model
mcp-mock Async MCP-style tool discovery and an agent tool loop None None
mcp-postgres Read-only database questions with MCP discovery and execution owned by the native UI client mcp_tools Upstream model, selected gateway, DBHub, and PostgreSQL
nested-subgraphs Streaming, status, human review, and background runs inside nested subgraphs, with no model call background, interrupts PostgreSQL checkpointer and run coordinator; Hatchet for background runs
multi-node-streaming Two sequential fake-model nodes contributing ordered text to one assistant message None None
response-outcomes Native refusal content and incomplete terminal responses None None
status-events Portable status updates for native client UI None None
streaming-long-mock A slow, predictable 100-sentence stream for checking streaming, Stop, and the returned history None None
persistent-plot-agent A tool-calling agent with an editable thread-scoped chart None Upstream model, Files API, and PostgreSQL store
interruptible-approval Durable choice-or-text human review before protected actions interrupts PostgreSQL checkpointer and run coordinator
simple-graph Streamed model output and discoverable runtime settings None Upstream chat model
simple-graph-external-tools Client-provided function tools returned as model tool calls None Upstream chat model
server-tool Installed package versions and OpenAI-compatible web search selected by the client None Upstream model plus SearXNG, Degoog, or upstream OpenAI search
lgos-rag Cited answers from the published documentation of the installed LGOS release, using embedding and BM25 search None Upstream chat and embedding models, and GitHub release archives

Graphs reach every model dependency through the selected gateway; see How Requests Flow.

The demo API opens its PostgreSQL runtime during application startup, so PostgreSQL must be available even when you call a provider-free graph. Start it with the demo API instructions.

persistent-plot-agent stores application data with a LangGraph Store. deep-research, interruptible-approval, background-interrupt, and nested-subgraphs checkpoint graph execution. None makes LGOS the owner of UI conversation history.

Background Responses from deep-research, background-mock, background-interrupt, and nested-subgraphs are polling-only and stored by Hatchet. They still do not persist a UI conversation or expose event replay.

Source Map

Most graph code is owned by the independent demo/api project:

  • demo/api/src/lgos_demo_api/registry.py registers graph names as OpenAI model names for lgos serve and lgos worker.
  • demo/api/src/lgos_demo_api/graphs/ contains its graphs and adapters.
  • demo/api-coding-agent/src/lgos_api_coding_agent/ owns the separately deployed coding-agent showcase, currently backed by Codex.

Continue with Run the Demo API for request examples or Chainlit and Open WebUI for UI behavior.