Each subdirectory is a standalone project — its own
pyproject.toml, .env (gitignored), and resources.yaml where
relevant. Copy any directory out of the repo and it just works.
cd examples/python/exNN_<name>
uv sync
uv run python main.py(Or pip install -e . + python main.py if you prefer plain pip.)
For LLM examples, copy .env.example to .env and fill in your key
before uv run.
| # | Example | Install | What it teaches |
|---|---|---|---|
| 01 | ex01_hello_world |
operonx (tier 1) |
@op, @graph, >>, START / END |
| 02 | ex02_data_pipeline |
operonx (tier 1) |
Linear pipelines |
| 03 | ex03_llm_chat |
operonx[openai] |
PromptOp / LLMOp / chat() |
| 04 | ex04_llm_advanced |
operonx[openai] |
Structured output, tool calling, multi-turn |
| 05 | ex05_loops_and_branches |
operonx (tier 1) |
Generator ops + if_() routing |
| 07 | ex07_embeddings_and_rag |
operonx[providers] |
EmbeddingOp + cosine RAG + optional reranker |
| 08 | ex08_error_handling |
operonx[openai] (subset of scenarios runs on tier 1) |
Capture, route, retry, LLM fallback |
| 09 | ex09_agent_workflow |
operonx[openai] |
Tool-calling agent on @graph.loop |
| 10 | ex10_multi_model |
operonx[openai] |
Parallel, routing, load-balance, fallback, ensemble |
| 11 | ex11_parallel_advanced |
operonx (tier 1) |
Fan-out / fan-in, generator iteration, partial failure |
| 12 | ex12_rag_advanced |
operonx[providers] |
Keyword RRF + hybrid (vector + keyword) RAG |
| 13 | ex13_graph |
operonx (tier 1) |
@graph composition + nesting |
| 14 | ex14_streaming_tracing |
operonx (tier 1) |
Generator pipelines, sync + async |
| 15 | ex15_callbot_streaming |
operonx (tier 1) |
Multi-level streaming pipeline |
ex06 is intentionally omitted (was a tracing tutorial; every example that uses an LLM now documents its own tracing setup).
Same as the top-level examples/README.md — tier 1
(operonx) is engine-only (no provider SDKs). [openai],
[anthropic], [gemini], [bedrock] add a single provider.
[providers] is shorthand for the embedding + reranker + multi-LLM
combo. [standard] and [all] are pre-bundled meta combos.
resources.yaml is per-example so each demo proves a real install
slice — no shared config, no umbrella inheritance. Provider examples
ship a minimal config naming only the resources that example uses.