llm-call/openrouter
OpenRouter inference with an explicit model and local secret reference. This is not the current CLI agent.
Inputs
| Field | Type | Contract |
|---|---|---|
| instructions * | string | templated. SPEC-MANIFEST.md §5.2. |
| format | string | text | json. Default text. |
| maxTokens | number | Default 1200. |
| temperature | number | Default 0.3. |
| role | string | prepended as "You are the ⟨role⟩ agent…" |
| model | string | overrides the loop's model. |
Outputs
| Field | Type | Contract |
|---|---|---|
| text | string | format: text — {text, model}. |
| model | string | The model that answered. |
| json | object | When format: json, the output is the parsed object instead of {text, model}. |
Behavior and limits
OpenRouter inference with an explicit model and local secret reference. This is not the current CLI agent.
Try it in context
Use the linked loops to inspect actual configuration and how output is passed to the next step. Keep credentials in your local environment.
Validation and run evidence
Run evidence not supplied. Release acceptance: pending.
No new validation or execution is performed by viewing this page.
No output artifact is available for this material.
Loops using this component
- Find your brand mentions
Filter a news feed for relevant mentions of your brand.
- Test the budget ceiling
Verify that a zero budget stops the loop before a paid step.
- Draft, check, approve, publish
Prepare a post and pause for your approval before publishing.
- Content feed
Read an RSS feed, write a short morning digest and send it to Telegram. Repeated runs may redeliver; receiver-side deduplication requires separate configuration.
- Feed fan-out
Give each entry its own model call and outgoing message.
- Get a structured digest
Request a JSON summary and reject incorrectly formatted output.
- Watch for price changes
Read a price source and prepare an update for your destination.
- Release watch
Follow a public GitHub repository and summarize what changed.
- Check before continuing
Review an output against a rubric before the next step.
Explore example uses
Illustrative scenarios based on these loops. These are ideas to adapt, not verified customer results.
- Your morning, distilled
A short morning brief with the updates that deserve your attention.
- From sources to a story
Turn selected sources into a cited story you can review before delivery.
- A second pair of eyes
A focused review of typography, contrast and keyboard access.
- Hear the useful mentions
Separate useful brand mentions from the noise in a selected feed.
- Keep up without keeping tabs
Know which repository updates need a closer look.
- An inbox that reads itself
Turn a source feed into a reading queue with enough context to choose.
- One brief. A better first draft.
A launch draft grounded in the product brief and its actual capabilities.
- Notice the price change
Bring a price observation into a brief you can act on.
- The conversations worth reading
Collect recurring questions and useful answers into one brief.
- Turn a reading list into context
Give each source a careful read, then build a referenced research note.
- A human at the right moment
Pause a workflow at the point where someone needs to decide.
- Less blank-page Monday
Shape a week of content around one brief and a clear audience.
- Find the signal in feedback
Find repeated needs and useful surprises in product feedback.
- Know when to stop
Make a spending boundary visible before a paid step.
- Notes your tools can use
Get a digest in a shape your next tool can read.
- A changelog for your team
Translate release notes into a useful update for your team.
- Publish after the yes
Keep the exact draft you approved tied to its delivery.
- A quieter daily routine
Bring routine updates together before you start the day.
Recorded examples
One prompt to your result
Your agent handles setup, configuration and checking the result in your project.