DEV Community: ai-coding-radar

Build a daily remote AI jobs feed with n8n

ai-coding-radar — Wed, 19 Aug 2026 04:22:16 +0000

Build a daily remote AI jobs feed with n8n

A useful job alert needs more than a keyword search. It should merge several
public feeds, remove duplicates inside the current result set, preserve the
original application and attribution links, and keep unavailable sources
visible instead of pretending that a partial search is complete.

This tutorial builds that narrow data pipeline with n8n and a public Apify
Actor. It collects public remote listings and returns normalized records. It
does not log in to job sites, scrape authenticated pages, or submit
applications.

Start with the public AI jobs example

The Remote Jobs Aggregator Actor
collects listings from Arbeitnow, Jobicy, Remote OK, and Himalayas. The
ready-to-copy daily AI and machine-learning jobs example
uses four related terms with any matching so one listing does not need to
contain every phrase.

For a no-code starting point, import the credential-free
n8n webhook workflow
and follow the n8n and Make recipes.
The source and input/output contract are public in the
repository.

Use a bounded, reviewable input

The public AI example uses this input:

{
  "sources": ["arbeitnow", "jobicy", "remoteok", "himalayas"],
  "keywords": [
    "machine learning",
    "artificial intelligence",
    "generative ai",
    "llm"
  ],
  "keywordMatchMode": "any",
  "locations": [],
  "maxAgeDays": 7,
  "limit": 50,
  "includeDescription": false
}

any means a record can match any one of the four phrases. Use all for a
strict query that must contain every keyword. The filter checks the published
title, company, tags, and location fields. Descriptions are disabled here so a
daily alert does not ingest a large block of untrusted listing text.

The Actor merges unique records round-robin across the selected sources after
filtering by age. This stops a high-volume feed from hiding every record from a
smaller source. The result limit is a cap, not a promise that 50 matching jobs
exist.

Call it from n8n without exposing a token

The imported workflow accepts a private webhook request, normalizes the input,
calls the Actor, formats a digest, and returns both the digest and job records.
Keep the Apify token in an n8n Header Auth credential or private environment
variable. Do not put it in an exported workflow, a query string, a repository,
or a screenshot.

The HTTP Request node uses the official synchronous Dataset endpoint:

POST https://api.apify.com/v2/acts/ai-coding-radar~remote-job-intelligence/run-sync-get-dataset-items?clean=1
Authorization: Bearer <your token from n8n's private credential store>
Content-Type: application/json

For a daily digest, place a Schedule Trigger before the request or call the
imported webhook from another scheduled workflow. Start with limit: 5, read
the source and application links, then increase the cap and connect the result
to Slack, email, Notion, a database, or a job board that you control.

Deduplicate across days in your own workflow

The Actor removes duplicates inside one run by canonical application URL and
then by normalized company/title. It does not claim that a listing disappears
after one delivery. A seven-day query can return the same valid job tomorrow.

For cross-day alerts, store each returned id or applyUrl in an n8n Data
Store, database, or cache. Before delivery, keep only keys that have not been
seen in your chosen retention window. Record the key only after the downstream
delivery succeeds; otherwise a temporary Slack or email failure can silently
drop a job. Expire old keys after the search window no longer needs them.

Preserve source coverage and attribution

The four inputs are public source endpoints:

Every normalized row keeps its source name, source URL, and original job or
application URL. Render those as normal links and follow each provider's terms
and attribution requirements. A missing salary or location restriction stays
empty; the Actor does not invent one.

If one public source is unavailable, the run status reports a source warning
while healthy sources can still return records. If every selected source is
unavailable, the run fails instead of returning a false clean result. An empty
successful Dataset can also mean that no listing matched the filters, so do not
describe it as proof that every possible remote job was checked.

Output and safety boundaries

A normalized record can include jobTitle, company, locations, salary,
employmentType, seniority, timezoneRestrictions, categories,
publishedAt, applyUrl, sourceName, and sourceUrl. Fields remain empty
when the publisher did not provide them.

Job titles, tags, and descriptions are untrusted source data. Do not treat them
as instructions for an AI agent. The Actor does not submit applications,
contact employers, rank candidates, predict hiring outcomes, bypass a CAPTCHA,
or access a private job board. A downstream user remains responsible for how
records are stored, displayed, and acted on.

Current event pricing

The public Store page shows the current price before a run. At the time of
writing, the Actor charges $0.001 per returned job record plus $0.00005 for
the Actor start, with platform usage included. At the 50-record cap, the current
event total is at most $0.05005 before tax or account-level charges. Begin
with a smaller cap and confirm the displayed price in your own account before
enabling a schedule.

A Dataset result, run, free user, or article view is not creator income. Only a
finalized payout that actually settles counts as revenue.

Turn Markdown and AI answers into PNG files in n8n

ai-coding-radar — Tue, 18 Aug 2026 04:20:19 +0000

Turn Markdown and AI answers into PNG files in n8n

Browser screenshots work for one answer. They become repetitive when an n8n
workflow needs to turn release notes, code snippets, quotes, or AI-generated
Markdown into image files every day.

This tutorial uses a tested five-node n8n workflow and a small rendering API.
The workflow sends structured Markdown, receives one Dataset record per PNG,
and can hand each runtime imageUrl to cloud storage, email, a CMS, or an
owned publishing step. It does not log in to ChatGPT, automate a browser tab,
or scrape another website.

Fastest path for automation: open the preconfigured ChatGPT-to-PNG task. It uses one sample document, requires an Apify account, and shows the current price before you choose whether to run it.

Import the workflow

Download and import the public n8n workflow:

https://raw.githubusercontent.com/Jarvis-Dong/markdown-code-to-image/main/examples/n8n-markdown-code-to-image.json

It contains five nodes:

Setup notes explains the credential and output contract.
Run manually provides a safe trigger for the first test.
Build batch input creates two Markdown documents.
Render Markdown and code calls the Actor's synchronous Dataset API.
Return PNG records passes the generated image records downstream.

The exported JSON contains no API token, credential ID, cookie, private
webhook, or signed output URL.

Configure authorization without leaking a token

Set APIFY_TOKEN in the private environment of your own n8n instance. The HTTP
node reads it at runtime and sends it as a Bearer authorization header. Do not
paste the token into the workflow JSON, a query string, a public repository, or
a screenshot.

The request uses this official endpoint:

POST https://api.apify.com/v2/acts/ai-coding-radar~markdown-code-to-image/run-sync-get-dataset-items?clean=1
Authorization: Bearer <your token from n8n's environment>
Content-Type: application/json

For production, replace the manual trigger with a Schedule Trigger, Webhook, or
an event from your CMS. Replace the static input node with Markdown from your
database, changelog generator, or AI step.

Send a valid batch

A request can contain 1–20 documents. Every item needs non-empty Markdown and
may have a short title:

{
  "documents": [
    {
      "title": "Release note",
      "markdown": "# Shipped\n\nThe useful change is live.\n\n```js\nconst ready = true;\n```"
    },
    {
      "title": "A concise AI answer",
      "markdown": "## Three steps\n\n1. Define the outcome.\n2. Automate repeatable work.\n3. Measure the result."
    }
  ],
  "theme": "paper",
  "width": 1080,
  "fontSize": 22,
  "watermark": ""
}

Available themes are paper, midnight, terminal, and clean. Width is
limited to 640–1600 pixels, body font size to 16–30 pixels, and each Markdown
value to 12,000 characters. Oversized or malformed input fails instead of
silently producing a partial image.

Use each returned PNG

The API returns an array. Each successful document has a record containing:

{
  "title": "Release note",
  "theme": "paper",
  "width": 1080,
  "height": 742,
  "format": "png",
  "markdownChars": 78,
  "storageKey": "IMAGE-001.png",
  "imageUrl": "https://runtime-generated-download-url.example",
  "generatedAt": "2026-08-15T08:00:00Z"
}

Treat imageUrl as a runtime value. Map it directly into an n8n HTTP Request
download, S3-compatible storage, email attachment, or another system you own.
Do not hard-code a previous run's URL; Apify storage retention follows the
caller's plan and settings.

Start from a focused use case

The ready-to-copy ChatGPT/AI-answer task shows a complete one-image input:

https://apify.com/ai-coding-radar/markdown-code-to-image/examples/chatgpt-markdown-answer-to-png

The public Actor and its general Markdown/code example are here:

https://apify.com/ai-coding-radar/markdown-code-to-image

For a one-off image that does not need an API, Cardify provides a free local
browser editor:

https://cardify.1222155.xyz/chatgpt-to-image/

The browser tool and the Actor have different jobs: use the browser editor for
manual previews, and use the Actor for scheduled, batch, n8n, or Make
automation.

Rendering and security boundaries

The renderer accepts Markdown, not arbitrary remote web pages. Raw HTML is
disabled, Markdown images are replaced with alt text, and the rendering page
blocks network requests. It does not fetch remote fonts, tracking pixels,
images, Mermaid diagrams, or authenticated pages. These limits prevent input
from turning the renderer into a request proxy.

The current output format is PNG. The generated file uses system fonts, and a
single rendered image is capped at 16,000 pixels high.

Cost

The public Store page shows the current price before a run. At the time of
writing, the price is $0.005 per generated image plus $0.00005 per Actor
start, with platform usage included. Begin with one document and confirm the
displayed cost in your own account before enabling a schedule.

A generated image, free user, page view, or test run is not creator income.
Only a finalized payout that actually settles counts as revenue.

Build a no-code UK supplier change monitor with n8n

ai-coding-radar — Mon, 17 Aug 2026 04:06:07 +0000

Build a no-code UK supplier change monitor with n8n

A supplier record can change long after onboarding. A company may change its
registered name, status, office address, SIC codes, accounts dates, or filing
details. Checking every record by hand does not scale, but a full credit-risk
platform can be excessive when the requirement is simply: tell me which
public register fields changed.

This tutorial builds a scheduled n8n workflow around a small Companies House
change feed. It watches only company numbers you provide, stores a baseline,
and forwards observed differences or source failures. It does not scrape a
website, predict insolvency, or make a credit or legal decision.

Fastest path: open the preconfigured supplier-monitoring task. It uses one public example company, requires an Apify account, and shows the current price before you choose whether to run it.

What the workflow does

The five-node workflow runs on a daily schedule:

Define a list of UK company numbers and a stable monitor ID.
Call the public Apify Actor through its official REST API.
Compare the latest basic register data with the named stored baseline.
Keep only changed, source_not_found, and source_error rows.
Hand those rows to a notification node that you choose.

The first successful check creates the baseline. Later healthy runs with no
differences return no alert rows by default. A source timeout or malformed
response remains visible as an error; it is never converted into "no change."

Import the tested n8n workflow

Download and import this workflow JSON:

https://raw.githubusercontent.com/Jarvis-Dong/uk-company-change-alerts/main/examples/n8n-uk-company-alerts.json

The exported workflow contains no API token, cookie, credential ID, private
webhook, or notification destination. After import:

Create an n8n Header Auth credential named Authorization with value Bearer YOUR_APIFY_API_TOKEN.
Attach that credential to the HTTP Request node.
Replace the sample company number with the suppliers you actually monitor.
Keep the same monitorId on future runs of the same watchlist.
Replace the final no-op node with Slack, email, Teams, Notion, or a private webhook, then activate the schedule.

The sample input is deliberately small and valid:

{
  "companyNumbers": ["02050399"],
  "monitorId": "supplier-watchlist",
  "emitUnchanged": false
}

Use a different monitorId for an independent watchlist. Reusing the same ID
for unrelated lists would mix their baselines.

Call the same feed from any automation tool

n8n is optional. The same input works with Make, a scheduled script, GitHub
Actions, or another system that can send an authenticated HTTP request:

curl -X POST \
  'https://api.apify.com/v2/actors/ai-coding-radar~uk-company-change-alerts/run-sync-get-dataset-items?clean=1' \
  -H "Authorization: Bearer $APIFY_API_TOKEN" \
  -H 'Content-Type: application/json' \
  -d '{"companyNumbers":["02050399"],"monitorId":"supplier-watchlist","emitUnchanged":false}'

Keep the token in your automation platform's credential store or an environment
variable. Do not place it in a public URL or exported workflow.

Understand the output before alerting

A first observation returns a baseline_created row. That is setup evidence,
not a warning. Later differences return one changed row per changed field:

{
  "companyNumber": "02050399",
  "companyName": "Example Company Limited",
  "observedAt": "2026-08-15T08:00:00Z",
  "sourceUrl": "https://data.companieshouse.gov.uk/doc/company/02050399.json",
  "status": "changed",
  "field": "companyStatus",
  "previousValue": "active",
  "currentValue": "dissolved",
  "errorCode": null,
  "errorMessage": null
}

The n8n filter forwards actual changes and explicit source errors. It ignores
baseline_created and healthy unchanged checks so the workflow does not send a
false alert during setup or create daily noise.

Data and decision boundaries

The Actor reads the official Companies House URI JSON source. You can inspect
the sample record directly:

https://data.companieshouse.gov.uk/doc/company/02050399.json

Companies House documents its public data products here:

https://www.gov.uk/guidance/companies-house-data-products

The monitored fields stay at company level: company name and status, country
of origin, incorporation or dissolution dates, registered office address,
previous names, accounts and return dates, mortgage counts, and SIC text. The
feed does not collect officer or PSC profiles, birthdays, private databases,
or derived risk scores.

Register data can still contain personal information, especially when a
registered office is also a residential address. Downstream users remain
responsible for access control, retention, redistribution, and the decisions
they make from the observations. observedAt means when this workflow saw the
value; it is not a guarantee that the source updates in real time.

Cost and source code

The ready-to-copy supplier monitoring task and public Actor are available here:

Its Store page shows the current pay-per-event price before a run. At the time
of writing, a successful company check is $0.003, an observed changed field
is $0.01, and the Actor-start event is $0.00005. Source failures and
rejected input are not charged as successful company checks. Apify platform
usage may also apply, so start with one company and confirm the displayed cost
for your own account before expanding a watchlist.

The source, n8n file, and Make setup notes are public:

https://github.com/Jarvis-Dong/uk-company-change-alerts

This is intentionally a narrow change-detection primitive. Treat its output as
an observed public-register difference that can trigger human review, not as a
credit report, legal opinion, compliance certification, or insolvency
prediction.

Build a fact-only Grants.gov opportunity monitor in n8n

ai-coding-radar — Sun, 16 Aug 2026 01:02:28 +0000

Build a fact-only Grants.gov opportunity monitor in n8n

Grant searches are useful only when the search result and its limitations stay
visible. This workflow watches a focused Grants.gov query and emits an
opportunity the first time it is seen, then emits it again only when a watched
summary field changes. It is a small change feed for an n8n, Make, or scheduled
script workflow; it is not an application service or a grant-writing tool.

Start with the public Actor and example

The tested Grants.gov Opportunity Monitor Actor
uses the official Grants.gov API Guide
and needs no Grants.gov login, API key, cookie, proxy, or browser automation.
The ready-to-copy Daily AI Federal Grant Opportunity Alerts example
shows a narrow query that can be scheduled after you inspect one result.
Running the Actor requires an Apify account, and the Store shows the current
price before you choose whether to run it.
For a no-code starting point, import the credential-free
n8n workflow
and follow its n8n and Make setup recipe.

The source code and input/output contract are public in the
Grants.gov monitor repository.
The Actor calls only the official unrestricted
search2 endpoint. It does not call
the detail or attachment endpoints, so it does not download long descriptions,
contact records, or application files that are not needed for a change alert.

Configure one stable monitor

Create an n8n HTTP Request node (or use the public example) with the following
JSON input. Keep one monitorId for one exact set of filters; use another ID
when the query changes so unrelated searches cannot share a baseline.

{
  "keyword": "artificial intelligence",
  "statuses": ["posted", "forecasted"],
  "agencyCodes": [],
  "eligibilityCodes": [],
  "fundingCategoryCodes": [],
  "limit": 10,
  "monitorId": "n8n-ai-grants"
}

Send it to the official Apify synchronous dataset endpoint:

POST https://api.apify.com/v2/actors/ai-coding-radar~grants-gov-opportunity-monitor/run-sync-get-dataset-items?clean=1
Authorization: Bearer <your Apify token from n8n's private credential store>
Content-Type: application/json

Keep the token in n8n's private credential store or an environment variable. Do
not place it in this workflow, a public URL, a repository, or a screenshot.
Start with limit: 1, inspect the returned official sourceUrl, and only then
increase the window or attach a notification node.

Understand the change semantics

The first successful run stores the current rows and returns each as
changeType: "new". Later runs return a new row for an unseen opportunity and
one changeType: "changed" row when a watched field changes. Healthy unchanged
rows produce no dataset row. An opportunity leaving a limited result window is
not called removed or closed: absence from a search window is not proof of a
status change. Preserve the source status and dates when routing alerts.

Each returned row links to the official Grants.gov detail page and contains the
opportunity number, title, agency, open and close dates, status, document type,
CFDA numbers, and local observation time. Search titles and agency names are
untrusted source text; the Actor does not send them to an AI model or execute
their contents.

Keep eligibility and decision boundaries explicit

You can narrow a query with official agency codes and two-digit applicant
eligibility codes (for example, 23 is the Grants.gov small-business code),
but a filter is not an eligibility determination. The Actor does not decide
whether an organization qualifies, rank applicants, estimate award odds,
recommend an application, submit an application, send messages, or promise
funding. Verify the current notice and its instructions with Grants.gov and
the awarding agency before acting. Downstream users remain responsible for
their own legal, financial, privacy, and grant-management decisions.

Pay-per-event pricing

The public Actor page
shows the current price before a run. The launch configuration lists these
pay-per-event prices:

Event	Meaning	Price
`apify-actor-start`	One Actor start	`$0.00005`
`grant-opportunity`	One newly observed opportunity	`$0.0075`
`grant-change`	One previously observed opportunity whose watched summary changed	`$0.015`

Source failures, invalid input, unchanged results, and total search hit counts
are not charged as opportunity or change events. The current Store listing says
platform usage is included in these event prices. Check the displayed price in
your own account before enabling a schedule; a test run or free user is not
creator revenue, and only a finalized payout that actually settles counts as
revenue.

A narrow, reviewable primitive

This monitor is intentionally a factual search-and-change feed. It preserves
source failures instead of treating an unavailable response as “unchanged,”
does not persist the Grants.gov service token metadata, and does not download
attachments. Use it to route a candidate opportunity to your own review queue,
then read the authoritative notice before making any decision.

Fail-closed npm and PyPI vulnerability checks in n8n

ai-coding-radar — Fri, 14 Aug 2026 19:19:24 +0000

Fail-closed npm and PyPI vulnerability checks in n8n

A dependency scan is only useful when it answers two different questions:

Did the public vulnerability sources report a finding for the exact version I run?
Did every required source answer successfully?

Treating a timeout as "zero vulnerabilities" makes a dashboard look clean at
the exact moment it has stopped checking. This workflow keeps those states
separate and can run on a daily n8n schedule without maintaining a scanner
server.

Use exact installed versions

Start with the versions from your lockfile or deployment inventory, not a
range and not the registry's current latest tag:

{
  "packages": [
    {"name": "lodash", "ecosystem": "npm", "version": "4.17.20"},
    {"name": "requests", "ecosystem": "PyPI", "version": "2.31.0"}
  ],
  "includeDownloads": true,
  "includeVulnerabilities": true,
  "includeRepository": true,
  "includeCisaKev": true
}

An exact version lets OSV answer for what is actually deployed. The result also
contains the registry's latest version, but it never silently replaces the
version you requested.

Import the workflow

Download the tested five-node n8n workflow:

https://raw.githubusercontent.com/Jarvis-Dong/oss-package-health-monitor/main/examples/n8n-oss-exact-version-scan.json

After import:

Create an n8n Header Auth credential named Authorization with value Bearer YOUR_APIFY_API_TOKEN and attach it to the HTTP Request node.
Replace the two sample packages with your exact installed versions.
Replace the final no-op node with Slack, email, Teams, Notion, a database, or a private webhook.
Test once, then activate the daily schedule.

The public workflow export deliberately contains no token, cookie, credential
ID, or destination URL.

Route findings without hiding failures

The filter forwards a row when any of these conditions is true:

vulnerabilities contains an OSV record;
cisaKevMatches contains a CVE listed in CISA KEV;
status is partial or error because a required source failed;
isLatest is false, so the installed version can be reviewed.

An OSV error leaves vulnerabilities and vulnerabilityCount as null. It is
not converted into an empty list. That distinction is what makes the workflow
fail closed instead of producing a false clean result.

The Actor queries public npm and PyPI registry metadata, the
OSV API, and the
CISA Known Exploited Vulnerabilities catalog.
It does not access private packages or automatically change dependencies.

Try the API with visible pricing

The public Actor and exact-version example are available here:

Actor: https://apify.com/ai-coding-radar/oss-package-health-monitor
Example: https://apify.com/ai-coding-radar/oss-package-health-monitor/examples/scan-installed-npm-and-pypi-versions-for-cves
Source and Make recipe: https://github.com/Jarvis-Dong/oss-package-health-monitor

The published price is $0.0015 per returned package record plus the small
Actor-start event shown on the Store page. Start with a short dependency list,
keep the source-status fields in downstream alerts, and treat the output as
point-in-time evidence rather than a security certification.