Examples
A categorized catalog of the runnable examples in the dynamiq repository — agents, orchestrators, RAG, tools, checkpoints, streaming, evaluations, and full use-case apps.
The dynamiq repository ships a large examples/ tree of runnable scripts. components/ demonstrates individual SDK features; use_cases/ contains end-to-end applications. Most scripts share the helper llm_setup.py, which builds an LLM node from a provider name (OpenAI, Anthropic, Cohere, Groq, or Gemini) — set the matching API key env var before running.
Agents
| Example | What it shows |
|---|---|
| simple_agent_wf.py | A minimal simple-agent workflow. |
| reflection_agent_wf.py | The reflection agent pattern — draft, critique, revise. |
| agent_wf.py | An Agent with tools inside a Workflow. |
| agent_multi_tool_workflow.py | One agent coordinating multiple tools (plus a streaming variant). |
| agent_e2b_sandbox.py | Agent with an E2B sandbox: remote files plus shell execution. |
| agent_daytona_sandbox.py | Same pattern on a Daytona sandbox backend. |
| agent_filesystem_interaction.py | Agent file-store usage: reading and writing files during a run. |
| use_agent_with_agent_tool.py | Agents as tools of other agents (sub-agents), with a memory variant. |
| use_agent_with_parallel_agent_tool.py | Parallel sub-agent tool calls. |
| use_subagent_checkpoint.py | Checkpointing a multi-node agent flow and resuming after a simulated crash. |
| use_agent_with_error_handling.py | ErrorHandling (timeouts, retries, backoff) on the LLM and the agent. |
| use_agents_vision.py | Agents over image inputs. |
| context_management_example.py | Agent context-window management. |
| use_agents_hidden_params.py | Hiding or requiring tool parameters with input_param_modes. |
| agent_vector_store_write_pipeline.py | Agent-driven vector-store writes (also as a writer tool). |
| use_neo4j_text2cypher_workflow.py | Text-to-Cypher agent over Neo4j. |
Orchestrators
components/agents/orchestrators/graph_orchestrator/
| Example | What it shows |
|---|---|
| code_assistant.py | Graph orchestrator coordinating a coding workflow. |
| concierge_orchestration.py | Routing between specialist agents. |
| email_writer.py | Multi-step drafting flow as a state graph. |
| trip_planner_orchestration.py | Multi-agent trip planning. |
| graph_orchestrator_yaml.py | Defining a graph orchestrator in YAML. |
See Orchestrators for the concepts.
Streaming
components/agents/streaming/ — per-agent-type streaming servers and clients (react/, simple/, reflection/), plus intermediate_streaming/ for step-by-step agent and orchestrator events. Lower-level transports live in components/core/websocket/: FastAPI WebSocket and SSE servers and Streamlit chat apps. Concepts: Streaming & Callbacks.
Human in the loop
components/tools/human_in_the_loop/
| Example | What it shows |
|---|---|
| confirmation_email_writer/console | Approval gates answered in the console. |
| confirmation_email_writer/socket | The same approval flow over WebSockets. |
| streaming_orchestrator | HITL feedback inside a streaming orchestrator. |
| streaming_post_writer | Streaming generation with human feedback events. |
Tools
| Example | What it shows |
|---|---|
| use_tavily.py, use_exa.py, use_serp.py | Web search tools. |
| use_firecrawl.py, use_jina.py | Scraping and content extraction tools. |
| use_python_node.py, use_http_api_node.py, use_sql.py | Python code, HTTP API call, and SQL nodes. |
| use_function_tool.py | Wrapping plain functions with function_tool. |
| use_react_with_coding.py, use_react_search.py, use_react_fc.py | Agents with code execution, search, and function-calling inference. |
| custom_tools/ | Custom Node tools: calculator, file reader, scraper-summarizer. |
| mcp_server_as_tool/ | Using MCP servers as agent tools. |
| pipedream/ | Pipedream-connected tools (Jira, files, configurable props). |
| stagehand_tool/ | Browser automation with Stagehand, including file upload flows. |
| cua_desktop/, e2b_desktop_sandbox/ | Computer-use and desktop sandbox automation. |
| multi_file_type_converter/ | Routing mixed file types through converters. |
Core: DAGs, YAML, checkpoints, tracing, cancellation
| Area | Highlights |
|---|---|
| dag/ | Building DAGs in code and loading them from YAML — LLM flows, fallbacks, MCP tools, agents with memory, skills, sandboxes, structured output. Pairs with YAML Workflows. |
| checkpoints/ | PostgreSQL-backed checkpointing: save per node, list, chain-walk, resume, cleanup. Pairs with Checkpoints, and see Worked examples for three complete inline crash-resume, HITL, and time-travel programs. |
| memory/ | Agent memory on every backend: in-memory, SQLite, PostgreSQL, DynamoDB, Pinecone, Qdrant, Weaviate, Dynamiq. Pairs with Memory. |
| tracing/ | Langfuse and AgentOps tracing handlers, plus flow visualization. Pairs with Tracing to Dynamiq. |
| cancellation/ | Mid-run cancellation: agents mid-loop, async tasks, YAML DAGs, HITL flows, with tracing. |
RAG
| Example | What it shows |
|---|---|
| vector_stores/pinecone_flow.py, elasticsearch_flow.py | Indexing flows into Pinecone and Elasticsearch. |
| vector_stores/filters/ | Metadata filtering at retrieval time. |
| vector_stores/delete_documents/ | Deleting indexed documents by file id. |
| retrievers/score_threshold_demo.py | Retrieval score thresholds. |
| rerankers/use_cohere.py | Reranking retrieved documents with Cohere. |
| embedders/embedders_execution.py | Running document/text embedders across providers. |
Also see components/splitters/ (character, token, semantic, code, HTML, JSON, markdown-header, contextual splitting) and components/helpers/converters/ (PDF, DOCX, PPTX, HTML, CSV, TXT converters). Concepts: RAG Pipeline and Document Processing.
LLMs
| Example | What it shows |
|---|---|
| llms/streaming.py, thinking_streaming.py | Token streaming, including reasoning streams. |
| llms/structured_output.py, function_calling.py | Structured output and tool/function calling. |
| llms/ollama.py, custom.py | Local and custom LLM endpoints. |
| llm_with_vision/ | Vision inputs and PDF extraction with vision models. |
| llm_with_files/ | Passing files to LLM nodes. |
Concepts: LLM Providers and Prompts & Messages.
Evaluations
components/evaluations/ — llm_evaluator.py (custom LLM-judged metrics), python_evaluator.py (programmatic metrics), workflow_eval.py (scoring a RAG workflow), and metrics/ with one script per built-in metric. Concepts: Evaluations.
Use cases
use_cases/ — end-to-end applications, most with a UI or server component:
| Use case | What it builds |
|---|---|
| gpt_researcher/ | A GPT-Researcher-style deep research pipeline, single- and multi-agent. |
| customer_support/ | Support agent backed by a mock banking API. |
| data_analyst/ | Data-analysis agent with a Streamlit front end. |
| erp_system/ | ERP assistant with a database tool, backend, and app. |
| financial_assistant/ | Financial Q&A assistant. |
| researcher/ | Research agent with app and backend. |
| trip_planner/ | Trip-planning agents with prompt templates. |
| job_posting/ | Job-posting generator. |
| literature_overview/ | Literature survey agent. |
| smm_manager/ | Social-media manager with a Mailgun tool. |
| project_manager/ | PM assistant with a Composio tool integration. |
| agents_use_cases/ | A grab bag of focused agents: coder, web researcher, deep scraping, feedback analyst, regression modeling, text-to-Cypher (Neo4j, Neptune, AGE), local and small LLMs. |
| chainlit/ | Chat UIs for Dynamiq agents with Chainlit. |
| graph_use_case/ | Graph ingest/query/check workflows. |
| search/ | Search app with server variants, including one served via Dynamiq. |
| agent_file_processing/ | File-processing agent behind an API server. |
Serving
cli/agent_service/ — a FastAPI service wrapping an agent, with Dockerfile, ready to deploy with the Dynamiq CLI. Pairs with CLI Overview and Deploy from the SDK.
Next steps
Sandboxes
Give agents an isolated remote filesystem and shell with E2B or Daytona sandbox backends — attach via SandboxConfig or drive the sandbox directly.
Worked Examples
Three complete, runnable checkpoint programs — crash-resume across a multi-node flow, a human-in-the-loop approval that survives a process exit, and time travel through a Graph Orchestrator's checkpoint chain.