Quickstart V2 examples

The numbered V2 demos are the fastest way to learn the shape of Cognis end-to-end. Each is a self-contained .rs file under examples/v2/. The LLM-backed demos (02, 03, 04, 06, 07, 08) read COGNIS_PROVIDER from the env — set it to ollama for local, or any of the other providers. Demos 01 and 05 are pure-Rust offline (no provider needed).

How to run

COGNIS_PROVIDER=ollama COGNIS_OLLAMA_MODEL=llama3.1 \
  cargo run -p cognis-examples --example 02_five_line_agent

The set

#	Name	What it shows	Source
01	`01_hello_runnable`	The smallest possible `Runnable<I, O>` impl — what your own chain step looks like.	src
02	`02_five_line_agent`	An agent in 5 lines: `Client::from_env` → `AgentBuilder::new` → `agent.run`.	src
03	`03_custom_graph`	Bypass `AgentBuilder` and hand-build the graph yourself, then wrap it as an `Agent`.	src
04	`04_streaming_chat`	Token-by-token streaming via `Client::stream` — what wires into a chat UI.	src
05	`05_checkpoint_resume`	Attach a `Checkpointer`, then time-travel through saved supersteps.	src
06	`06_ollama_tool_calling`	Calculator tool + arithmetic question — the canonical tool-calling agent loop.	src
07	`07_ollama_structured_output`	`StructuredOutputParser<T>` recovering a typed `Recipe` from free-form LLM text.	src
08	`08_ollama_multi_agent`	Sequential planner → executor handoff for a real task.	src

Recommended order

Read in number order for the full guided tour. If you’re impatient, jump to whichever maps to your use case:

Building a chat agent? → 02, then 06 for tools, then 07 for structured output.
Building a graph workflow? → 01, then 03, then 05 for checkpoints.
Building multi-agent? → 02, then 08.
Building a streaming UI? → 04, then Patterns → Streaming UI.

Quickstart guide

The narrative walkthrough of demo 02.

Examples → Agents

More agent patterns beyond the V2 set.

Examples → Graphs

More graph patterns beyond demo 03.

Patterns

Larger applications that combine multiple V2 demos.

​How to run

​The set

​Recommended order

​See also