AI Examples
Chat completions with OpenRouter and a secret-managed key, DAG-level defaults with a custom endpoint, response reuse, typed answers, sessions, extended thinking, workflows as tools, model fallback, and typed decisions that route on confidence. Every example runs as-is with an OPENROUTER_API_KEY exported. Cards that omit secrets and llm assume the setup block from the first card.
Provider Setup with a Secret
secrets:
- name: OPENROUTER_API_KEY
provider: env
key: OPENROUTER_API_KEY
llm:
provider: openrouter
model: deepseek/deepseek-v4-flash
steps:
- id: ask
action: chat.completion
with:
prompt: |
What is 2+2? Reply with just the number.The secrets entry resolves the key at run time and masks it in logs. The DAG-level llm block is inherited by every chat step that sets no LLM fields of its own.
Endpoint and Defaults at DAG Level
llm:
provider: openrouter
model: deepseek/deepseek-v4-flash
base_url: https://openrouter.ai/api/v1
api_key_name: OPENROUTER_API_KEY
system: |
Answer in one short sentence.
temperature: 0.2
max_tokens: 200
steps:
- id: ask
action: chat.completion
with:
prompt: What does a DAG scheduler do?base_url points at any OpenAI-compatible endpoint (shown with OpenRouter's own URL made explicit; the same field targets vLLM, Ollama, or LM Studio or a corporate proxy), api_key_name picks the environment variable holding the key, and system plus the sampling fields become defaults for every chat step. The full field list has the rest.
Use the Response in a Later Step
steps:
- id: ask
action: chat.completion
with:
prompt: |
What is 2+2? Reply with just the number.
output: ANSWER
- id: use_answer
run: echo "The model said ${ANSWER}"
depends: askoutput captures the completion text as a variable for downstream steps.
Typed Answer as Step Outputs
steps:
- id: classify
action: chat.completion
with:
prompt: |
Classify this customer note and extract the amount:
I was charged twice, please refund the extra 12.50 EUR.
output_schema:
type: object
properties:
category:
type: string
enum: [refund, complaint, question]
amount:
type: number
required: [category]
- id: record
run: echo "${steps.classify.outputs.category} ${steps.classify.outputs.amount}"
depends: classifyThe model answers through a tool whose parameters are the schema. The answer is validated, and each listed property becomes a step output.
Multi-turn Session
steps:
- id: ask
action: chat.completion
with:
prompt: |
What is 2+2? Reply with just the number.
- id: follow_up
action: chat.completion
with:
prompt: |
Multiply that by 3. Reply with just the number.
depends: askChat steps inherit the conversation from the steps they depend on, so "that" resolves to the earlier answer.
Extended Thinking
steps:
- id: reason
action: chat.completion
with:
provider: openrouter
model: deepseek/deepseek-v4-flash
thinking:
enabled: true
effort: low
prompt: |
A bat and a ball cost 1.10 in total. The bat costs
1.00 more than the ball. How much does the ball
cost? Reply with just the amount.thinking maps to the provider's reasoning controls; raise effort for harder problems.
Workflows as Tools
steps:
- id: ask
action: chat.completion
with:
provider: openrouter
model: deepseek/deepseek-v4-flash
tools:
- calculator
prompt: |
What is 15 times 23? Use the calculator tool,
then reply with just the number.
---
name: calculator
description: Multiply two numbers.
params: "a b"
steps:
- id: multiply
run: echo $(($1 * $2))Each name in tools exposes a DAG as a callable tool; its params become the tool's argument schema, and each call is a real child run. The explicit provider and model are required here: setting any LLM field under with (such as tools) replaces the DAG-level llm block instead of merging with it.
Model Fallback
steps:
- id: summarize
action: chat.completion
with:
model:
- provider: openrouter
name: deepseek/deepseek-v4
- provider: openrouter
name: deepseek/deepseek-v4-flash
prompt: |
Reply with the single word "ready".An ordered model list tries the next entry after retries for the current one are exhausted.
Agent Workflow
type: agent
llm:
provider: openrouter
model: deepseek/deepseek-v4-flash
steps:
- name: disk
description: Show filesystem usage.
run: df -h
- name: load
description: Show uptime and load average.
run: uptime
tasks:
- name: checked
description: Finished when both disk and load have been checked.Steps become a catalog of actions and tasks state the goals; the model decides what runs next. Built up example by example on the agent examples page.
Route on a Typed Decision
type: graph
steps:
- id: classify
action: decision.evaluate
with:
provider: openrouter
model: typesafe/jev-1.13
state: Please refund my duplicate charge.
questions:
refund:
type: noul
instructions: Is the customer asking for a refund?
- id: triage
action: router.route
with:
value: ${classify.output.answers.refund.noul}
routes:
"num:>=0.9": [auto_approve]
"num:<=0.1": [auto_reject]
depends: classify
- id: auto_approve
run: echo "Refund approved"
- id: auto_reject
run: echo "Not a refund request"
- id: human_review
preconditions:
- condition: ${classify.output.answers.refund.noul}
expected: "num:<0.9"
- condition: ${classify.output.answers.refund.noul}
expected: "num:>0.1"
run: echo "Needs a human"
depends: classifyA noul question answers yes or no as a probability, so the confident ends route automatically and the middle band goes to a person. A route carries one pattern, so the middle band states both bounds as preconditions instead.