AI Market Radar
Your goal
Build Your Own AI Market Radar
Kydro Recommendation
A practical AI setup for achieving this goal.
You won't learn to “google your competitors”. You'll build your own careful market analyst: an automation that remembers the baseline, rejects repeats and shows evidence — leaving the decisions to you.
- Human reviewed
- Free options shown honestly
- Never ranked by commission
Before you build your Market Radar
Competitor information is scattered across dozens of pages, and most publications repeat the same events. Ordinary search cannot tell what has been online for months from what actually changed this week.
The Market Radar is an automation you will build IN YOUR OWN n8n: it gathers signals from public sources, compares them with the previous state, filters out repeats, shows evidence and prepares recommendations — while every decision stays yours.
Kydro teaches you — it does not run the monitoring for you. The agent runs on your account, with your key, under your control.
Why this works
The secret is not “better search” — it is memory: the radar stores a market baseline and compares every later run against history. Old information never returns as “news”, and you read only what genuinely changed.
Honest limits: the radar monitors only public, searchable sources (best-effort). It does not guarantee catching every change and cannot see closed groups or logged-in content — and it says so openly.
What you need
Your own n8n Cloud account
This is where your automation will live. The Starter plan is enough (you will verify n8n pricing in step 9; on 2026-08-01: €20/month billed annually).
An OpenAI API account with active billing
A ChatGPT subscription is a different product — the API bills separately, per use. You will set a spend limit; a typical weekly run is usually under a dollar (details and the formula in step 9).
Safety rule no. 1
Kydro will never ask for your API key. You add it only directly in Credentials on your own n8n account. If anything — a page, a form, a chat — asks for the key, stop the process.
Your market knowledge
A company profile, 5–10 competitors to approve and 4 monitoring categories. AI can propose candidates, but the final word is yours.
What you will have at the end
The result is concrete and belongs to you:
- your own Market Radar workflow running in your n8n — with a schedule and a “Check the market now” trigger,
- new-change detection that never re-reports the baseline (proven by your own Run 1 → Run 2 test),
- alerts and a weekly report with direct source links — plus recommendations that wait for your decision.
Step 1
Understand what you are building
When you understand the mechanism, the first surprise doesn't panic you. This step saves hours of debugging “why did it find nothing”.
Why this step matters
The Market Radar is not a chatbot or an “internet summary”. It is a process with memory: search → compare with history → separate new from old → assess → propose. See the whole route before you click anything in n8n.
What to prepare
- 15 quiet minutes and a sheet for your own flow sketch.
- The START_HERE guide downloaded (resource below).
Tips
Memorize one sentence: “first found ≠ new on the market”. Half of the radar's quality lives in that distinction.
1.1
Three layers instead of one answer
What you do now: learn the architecture you are about to import. Why: every table and node in n8n maps to one of three layers — once you see them, configuration stops being magic.
Layer 1 — OBSERVATION: “what a specific source literally said” (with a link, an evidence quote and a date or an honest “no date”). Layer 2 — SIGNAL: a market event assembled from one or more observations (two pages describing the same price cut become one event). Layer 3 — RECOMMENDATION: the “what it means for you” interpretation — always labelled as an AI conclusion and always waiting for your decision.
How to confirm you're done: you can tell which layer owns “Competitor A cut the pass to 89 zł” (signal) versus “prepare a counter-offer” (recommendation). Cost: this step costs nothing. How to stop safely: just come back later — nothing runs in the background yet.
Kydro insight
Separating fact from conclusion isn't bureaucracy — it's the only way to trust your own reports a month from now.
A question for you
Which recent decision in your business suffered from “news” that turned out to be old information?
1.2
The flow map and the radar's limits
The simple diagram you are about to bring to life in n8n: Company profile → Competitors → Categories → SEARCH (AI queries the public index) → BASELINE (the first run stores everything as the reference point) → NEXT RUN → DEDUPLICATION (rules + semantic comparison) → IMPACT ASSESSMENT → REPORT or ALERT → YOUR DECISION.
Limits worth knowing from day one: the radar sees only what is public and indexed (no closed groups, no logged-in content, no full review databases); some pages carry no dates — such findings go to “needs verification” instead of alerts; AI can be wrong — which is why conclusions are labelled and conflicting sources wait for a human.
Common mistake
Treating the radar as an oracle. It is a careful analyst with evidence — it feeds your market knowledge, it doesn't replace it.
A question for you
Which source of competitor knowledge is beyond a search engine today — and who in your company could cover it manually?
Kydro Resources
You can continue if…
- I can describe each of the three layers in one sentence.
- I know what the radar will NOT see — and accept it consciously.
Done when: you can explain the three layers (observation → signal → recommendation) in your own words, and why a baseline is not “news”.
Step 2
Prepare accounts and safe credentials
One well-kept key is the difference between calm automation and a surprise bill. This step lays the security foundation.
Why this step matters
Four concepts get mixed up most often: a ChatGPT subscription (chat for humans), an OpenAI API account (a pay-per-use service for programs), the API key (a private technical password) and n8n Credentials (the safe where the key lives). Untangling them takes 5 minutes and saves weeks of frustration.
What to prepare
- An e-mail address and a card for OpenAI billing (you'll set a limit right away).
- A password manager — for account passwords. The API key is saved nowhere except n8n.
Tips
Name the key “n8n-market-radar” — when you ever revoke it, you'll know which one it is.
Nobody in this process should ever ask you to send your API key. If you see such a request, stop the process.
2.1
n8n Cloud — your automation's home
What you do: create an account at n8n.io (“Get started”), pick the Starter plan and sign in to your workspace. Why: n8n runs a “workflow” — a chain of blocks (nodes) executing actions in order; Cloud means it works without your computer being on.
What should appear: an empty workspace with a “Create workflow” button. How to confirm: you can see your address like your-name.app.n8n.cloud. If it looks different: make sure you didn't land on self-hosted — this guide uses Cloud. Cost: the n8n subscription per its pricing (step 9). How to stop: the account can wait — nothing starts by itself.
A question for you
Which repetitive task at work would you hand to an automation like this next?
2.2
The OpenAI API and the key — once, and only into n8n
What you do: create an API account at platform.openai.com, add a card under Billing and SET A MONTHLY LIMIT (e.g. $10), create an “n8n-market-radar” key under API keys and copy it ONCE. Why: the API is the connector that lets n8n send a job to OpenAI and receive the result — no programming on your side; the key is that connector's private password.
Now the safe: in n8n open Credentials → Add credential → choose “OpenAI” → paste the key → Save. What should appear: a green save confirmation. How to confirm: the credential is on the list and the key exists NOWHERE else (you may clear the clipboard). If it looks different: don't park the key “for a moment” in notes — go back and paste it directly. Cost: adding the key costs nothing; you pay only for runs. How to stop: if anything worries you, revoke the key in OpenAI and create a new one — that's the safe reset.
Optional (for e-mail alerts): prepare Gmail (the demo path) or SMTP (the alternative) — you'll connect it in step 8. The alert recipient is configured only in n8n; Kydro never collects it.
Kydro insight
A ChatGPT subscription does NOT include the API. They are two separate products with two separate bills — mixing them up is this step's most common false start.
Common mistake
Pasting the key into a chat, a “for later” note or a file. The key has one home: Credentials in your n8n.
A question for you
Where have your technical passwords ended up so far — and what will you change after this step?
Kydro Resources
Common mistakes
No spend limit set in OpenAI.
Why it hurts: A configuration mistake can then cost real money instead of stopping at a threshold.
How to avoid it: Set the limit (e.g. $10/month) BEFORE the first run; raise it deliberately later.
One API key for everything.
Why it hurts: You can't tell what generates cost and can't revoke the key without breaking other tools.
How to avoid it: A separate, project-named key — easy to identify and easy to cut off.
You can continue if…
- The key exists only in n8n Credentials.
- The OpenAI spend limit is set.
- I know a request for the key = STOP.
Done when: you have an n8n account, an OpenAI account with billing and a spend limit, and the API key exists only in your n8n Credentials.
Step 3
Import the setup and create the Data Tables
Data Tables are the radar's memory — without them there is no baseline and no dedup. The setup builds them for you and says plainly when something doesn't match.
Why this step matters
An n8n Data Table is a simple table (like a sheet) living in your instance. The radar uses them as memory: what it saw, when, with which content fingerprint. The setup workflow compares your instance with the manifest and creates ONLY the missing tables.
What to prepare
- The market-radar-setup.workflow.json file downloaded (resource below).
- n8n open in a second tab.
Tips
The setup is safe to re-run: matching tables are skipped, a mismatched schema STOPS the process with a diff list — it deletes nothing.
3.1
Importing your first workflow
What you do: in n8n click “Create workflow” → the ⋯ menu (top right) → “Import from file” → choose the downloaded market-radar-setup.workflow.json. Why: the import loads a ready node chain — you build nothing by hand. What should appear: a canvas of connected blocks (incl. “Manifest tabel”, “Lista istniejących tabel”, “Raport setupu”).
How to confirm: no node glows red. If it looks different: a red node usually means a newer/older n8n version — open it and confirm the fields; for trouble, use step 9's troubleshooting. Cost: zero — the setup never calls OpenAI. How to stop: close the tab; a workflow does nothing until you run it.
A question for you
What plays the role of “market memory” in your company today — and who updates it?
3.2
Running the setup and reading its report
What you do: click “Execute workflow” (a manual start) and wait a few seconds. Why: the setup READS the table list first, then creates only what's missing — that's why a second run breaks nothing (that property is idempotency: same operation, same effect, no duplicates). What should appear: the final node reporting “Setup OK: N matching, M created” and the full kydro_g7_… table set in your Data Tables list.
If it looks different: an “INCOMPLETE — schema mismatch” report names the table and missing columns; compare with the manifest (resource below) and add the column OR delete YOUR empty table and re-run. The setup itself deletes nothing. Cost: zero. How to stop: n8n's Stop button; a re-run finishes only what's missing.
Kydro insight
The manifest is a contract: the setup, the main workflow and the checklists all read the same table and column names — change things in the manifest, never “by eye”.
A question for you
Which competitor data would you love to browse in a table today?
Kydro Resources
You can continue if…
- Setup report: Setup OK.
- A second setup run = zero changes.
Done when: the setup ends with a “Setup OK” report listing all tables, and re-running it duplicates nothing.
Step 4
Import the Market Radar
This is the engine room: the main workflow, the report and the prompts. After this step the radar is complete — waiting only for your data.
Why this step matters
The main workflow ships with safeguards you don't see at first glance: retry (repeat on error), checkpoints (progress saved after every task), a parallel-start lock (a renewing lease) and a budget counter. This step shows where they live — so you trust them knowingly.
What to prepare
- Downloaded: the main + report workflows, PROMPTS.md and the three SO schemas (resources below).
- Step 3 finished (the tables exist).
Tips
Model names verified in the official OpenAI pricing docs on 2026-08-01: gpt-5.6-luna (extraction/classification) and gpt-5.6-terra (impact analysis). If your account doesn't list them, pick the current equivalent from the node's model list and note the difference — never guess names.
4.1
Import and credential binding
What you do: exactly like step 3, import market-radar-main.workflow.json and then market-radar-report.workflow.json (the report is a sub-workflow — the main one calls it after every run). Then, in each “KROK A/B/C1/C2” HTTP node, select your OpenAI credential from the list. Why: the JSON carries only the credential TYPE name — never a key value — so after import you point it once at your own safe.
What should appear: credential fields stop being empty and no node glows red. How to confirm: the KROK A node shows your credential and a model. If different: a missing “OpenAI” type on the list means step 2 wasn't finished. Cost: zero until you run something. How to stop: Save the workflow and return later — nothing starts on its own.
A question for you
Which engine-room safeguard (retry, checkpoint, lock, budget) gives you the most peace of mind — and why?
4.2
Prompts, SO schemas and the model
What you do: open PROMPTS.md and paste the prompts into their nodes (P1→KROK A, P2→KROK B, P3→KROK C1, P5→KROK C2, P6→report); into B/C1/C2 also paste the matching SO schema (the “schema” field of text.format). Why: a Structured Output is a “rigid form” for AI — the answer MUST match the schema, so a date cannot be invented and an off-list URL fails validation.
The model: set gpt-5.6-luna in A/B/C1 and the report, gpt-5.6-terra in C2 (or your list's equivalents; see the tip above). Retry: in every HTTP node, Settings → Retry On Fail (2 tries). What should appear: a saved workflow with no red nodes. How to confirm: previewing KROK B shows the so1 schema inside the request body. Cost: still zero. How to stop: Save and take a break — configuration doesn't run away.
Kydro insight
The division of labour: n8n rules count and guard (hashes, statuses, budget, alerts) while AI reads and writes (extraction, comparison, impact, report prose). Nothing that can be computed deterministically is handed to the model.
Common mistake
Skipping the SO schemas (“it works without them”). Without strict schemas, the first oddly-formatted reply corrupts your tables.
A question for you
Where in your work would you prefer answers on a “rigid form” instead of free text?
Kydro Resources
You can continue if…
- Both workflows imported and saved.
- Every AI node has a credential, model, prompt and (B/C1/C2) a schema.
- Retry configured on the HTTP nodes.
Done when: both workflows (main + report) are imported, bound to your OpenAI credential, with the model set and the SO schemas pasted.
Step 5
Configure your market
A radar is only as good as its target. A precise profile, approved competitors and 4 categories are the difference between a noise report and a decision report.
Why this step matters
There are four starter categories, deliberately narrow: (1) prices & promotions, (2) new locations, (3) new services or classes, (4) meaningful offer changes. Fewer categories = less noise and lower cost; widen them once you trust the results.
What to prepare
- The outputs of both resource prompts (profile + candidates).
- Fifteen minutes to consciously approve the competitor list.
Tips
FitNova from the examples is a FICTIONAL company — demo only. When entering real competitors, stick to facts; the radar never fabricates information about real businesses.
5.1
Profile, competitors and your approval
What you do: open the start form (the “onboarding” Form Trigger in the main workflow exposes a URL — it's inside the node) and fill the profile with prompt #1's fields; then enter candidates into the competitors table and SET status approved only for those you truly want to monitor (5–10). Why: monitoring will not start without at least one approved — a deliberate latch: no tracking “by momentum”.
Then: tick the 4 starter categories, set the run's max budget (default 60 search actions — explained in step 9) and enter the report recipient IN N8N (the company_profiles table). What should appear: one profile row and a competitor list with statuses. How to confirm: the approved count is within 5–10. If different: an over-long list raises cost and noise — trim deliberately. Cost: zero (still configuration). How to stop: the table data simply waits.
Kydro insight
AI proposes — you approve. That order isn't ceremony: it's your control over where budget and attention go.
A question for you
Which of your competitors changes fastest — and how do you notice it today?
Kydro Resources
You can continue if…
- 5–10 competitors with status approved.
- Budget and report recipient set in n8n.
Market Profile
One place for your configuration decisions: whom you watch, in which categories, on what budget. You paste its values into n8n — and revisit it whenever the scope changes.
- Company & segment
- Competitors (approved)
- Categories & budget
Done when: the company profile is saved, 5–10 competitors are approved, categories and budget set, the report recipient entered in n8n.
Step 6
Run 1 — the baseline
The baseline is the whole system's reference point: a photo of the market “as of today”. Without it, every later run would shout about old things.
Why this step matters
A correct Run 1 usually sends NO alert — and that is success, not failure. Everything it finds is the “existing state”: the knowledge from which changes start counting. An alert from Run 1 would mean the system confuses history with novelty.
What to prepare
- Steps 2–5 finished.
- The OpenAI spend limit set (step 2) — your safety net.
Tips
The first run tends to be the priciest (baseline ×1.5–2 of a normal run). Tens of cents to ~$2 for 5 competitors is the normal ballpark (an estimate; formula in step 9).
6.1
Start, watch, stop safely
What you do: open the “Check the market now” form (the main workflow's Form Trigger), confirm the start and watch the execution in the Executions tab. Why manual: you want to SEE the first run live — the schedule comes in step 8. NOTE — COST: from this moment paid OpenAI calls run (web search + tokens), counted live in monitoring_runs.
What should appear: competitor × category tasks moving pending → searching → done; after a few minutes the run finishes as completed/baseline. How to confirm: walk the post-Run-1 checklist (resource below). If different: a single failed task with a note isn't a disaster — retry often completes it; see step 9. How to stop SAFELY: the Stop button — checkpoints keep finished tasks, so resuming (starting again within the same window) completes only what's missing, with no duplicates.
Peek into the tables: source_observations (what was found and where, with the evidence quote), market_signals (status baseline), monitoring_runs (counters: API calls, web_search actions, tokens, time, cost ESTIMATE).
Kydro insight
Run 1 doesn't judge or alarm — it PHOTOGRAPHS. Judgement begins only once something exists to compare against.
Common mistake
Panic that “there are no alerts”. In Run 1 their absence is proof of correctness, not failure.
A question for you
What in the recorded baseline surprised you most — what didn't you know about your own market?
Kydro Resources
Common mistakes
Enabling the schedule before the first manual run.
Why it hurts: You lose the one chance to watch the system live and catch a config mistake cheaply.
How to avoid it: The schedule waits for step 8 — after the Run 2 test.
You can continue if…
- Baseline run = completed; the post-Run-1 checklist passed.
- I know what the run cost (the counters' estimate).
Done when: the baseline run is completed, the tables are filled, and the post-Run-1 checklist passes in full.
Step 7
Run 2 — verify the deduplication
This is the radar's most important quality test: old information MUST NOT return as new. A passed Run 2 means the reports can be trusted.
Why this step matters
You may execute Run 2 shortly after Run 1 — no need to wait for “real” market changes. The goal isn't to find news; it's proof that the baseline does NOT come back. No new changes is a correct result worth celebrating.
What to prepare
- Run 1 finished with status completed.
- The post-Run-2 checklist at hand (resource below).
Tips
The mechanics underneath: identical page content = a “visit without change” (only last_seen_at bumps); different URLs of one event merge into ONE signal with several evidence rows; changed content = an update (the old signal gets superseded_by); no date = undated_discovery, never an alert.
7.1
The second pass and reading the verdict
What you do: trigger “Check the market now” again (a normal run's cost — see step 9) and open market_signals afterwards. Why: the system just compared fresh findings against memory — every row received a status separating “new to the system” from “new on the market”: baseline / new_market_event / event_update / duplicates (exact and cross-source) / republished_old_information / undated_discovery / needs_verification.
How to confirm: walk the post-Run-2 checklist point by point (zero baseline shown as news; unchanged pages only bumped last_seen_at; conflicts marked contradicted). If different: a single false “new” usually means an undated page — check date_confidence and the evidence; step 9's troubleshooting continues from there. Cost: a standard run. How to stop: as always, Stop + safe resume.
Kydro insight
“No new changes” in a report is a first-class result — the radar is telling you: focus on your business, the market hasn't run away.
Common mistake
“Improving” prompts so Run 2 “finally finds something”. That's breaking the test, not improving the system.
A question for you
How many hours a month do you win back by not checking competitor pages “just in case”?
Kydro Resources
Common mistakes
Treating duplicates as a system error.
Why it hurts: Rejected duplicates are the job done well — their counter shows how much noise you were spared.
How to avoid it: Read the duplicate counter as a measure of value, not a defect.
You can continue if…
- Zero baseline items shown as new.
- I understand: duplicate / update / undated / contradicted.
Done when: Run 2 surfaced no baseline item as news, and the dedup checklist passes in full.
Step 8
Enable alerts, the report and recommendations
Here the radar starts WORKING for you: a weekly Market Radar in your inbox, an urgent alert only for truly important events, and recommendations awaiting your decision.
Why this step matters
AI NEVER decides to send an alert — rules do: a new event/major update + a credible, adequate source + a fresh date + high impact + no unresolved conflict + not undated + not sent before. Cap: max 3 alerts a week; the same alert never goes out twice (alerts_log).
What to prepare
- Gmail (demo) or SMTP (alternative) ready to connect in n8n.
- The model examples (resource below) for comparison.
Tips
With no recipient configured the system builds an alert PREVIEW and stops delivery — it never sends “into the dark”.
8.1
Schedule, e-mail and the alert test
What you do: set the day and hour in the Schedule node (e.g. Monday 8:00) and switch the workflow to Active; sign in to the Gmail node with your account (or fill SMTP) — the data stays in YOUR credentials. Why: from now on the radar performs the weekly run and mails the report itself; “Check the market now” still works in parallel (the lease lock keeps two runs from colliding).
The alert test: the preview shows the full structure — what happened, at which competitor, when, evidence and source, why it matters (labelled as interpretation), the confidence level, a recommendation with a deadline and “what needs verification”. The weekly report: a one-page Market Radar (rule-counted numbers, Top 5, opportunities/threats, 3 actions, a “Needs verification” section, direct links; an empty week = the explicit “no new meaningful changes” line). Recommendations: in the recommendations table you set accepted/rejected — your decision is stored in user_decisions; NOTHING executes itself.
How to confirm: the schedule shows Active and the alert preview matches the examples. If different: a missing e-mail usually means an unfinished Gmail/SMTP credential — step 9. Cost: the weekly run per step 9's formula. How to stop: switch the workflow to Inactive — a full, reversible schedule stop.
Kydro insight
An alert should be rare. Arriving daily isn't vigilance — it's noise; the rules and the 3/week cap protect your attention.
A question for you
Which competitor change truly deserves to interrupt your day — and which can wait for Monday's report?
Kydro Resources
You can continue if…
- The workflow is Active with a weekly schedule.
- The alert preview matches the model; I know where I accept recommendations.
Done when: the weekly schedule is on, e-mail (Gmail/SMTP) connected, an alert tested, and you know how to accept or reject a recommendation.
Step 9
Master costs, safety and troubleshooting
An automation whose costs and failures you don't understand will surprise you eventually. This step turns surprises into procedures.
Why this step matters
Prices verified in official sources on 2026-08-01 — re-check when they change: n8n Starter €20/month (annual billing); OpenAI web search $10 per 1,000 calls + search-content tokens at model rates; gpt-5.6-luna $0.20/$1.20 per 1M tokens (in/out); gpt-5.6-terra $2/$12. The key distinction: one Responses API call may perform SEVERAL web_search_call actions — the counter in monitoring_runs counts actions from the API response, it doesn't guess.
What to prepare
- The monitoring_runs table open, with your two runs.
Tips
The estimate formula: cost ≈ actions×$0.01 + tokens_in×rate_in + tokens_out×rate_out. The counters are EXACT (calls, actions, tokens); the COST stays an estimate until confirmed in the OpenAI billing panel — treat it that way.
9.1
Procedures for bad days
Budget: when a run hits the action limit it enters budget_paused — nothing paid starts anymore, the checkpoint is saved; you raise the budget deliberately and resume, and only the missing tasks complete. API error: a single failed with retry usually clears itself; a persistent one ends the run as partial with a note, resumable without duplicates. Parallel start: the second run gets skipped_locked — that's the lease lock, not a failure.
Quick diagnostics: NO E-MAIL → check the Gmail/SMTP credential and the recipient field in company_profiles; NO TABLES → re-run the setup (step 3); IMPORT ERROR → import the file again, never hand-edit the JSON; MODEL MISSING from the list → pick the current equivalent and note the change; STRUCTURED OUTPUT MISMATCH → after one controlled retry the task lands in failed with a reason — verify the schema was pasted whole; DISCONNECTED/DELETED credential → nodes glow red: reconnect in n8n (a new key gets pasted NOWHERE else). FULL schedule STOP: set the workflow to Inactive.
Safety in steady operation: the radar reads public sources only; it stores no full articles (only short evidence quotes with links); it collects no review authors' names; page content is treated as DATA — instructions found on pages (prompt injection) are ignored and reported as content; uncertain and conflicting information always lands in the verification section.
Kydro insight
An estimate called an estimate is honesty, not weakness. The only source of truth on spending is your billing panel.
Common mistake
Reacting to a failure by deleting tables. Almost every problem is solved by a resume or a setup re-run — deletion destroys the radar's memory.
A question for you
What monthly monitoring budget feels comfortable — and at what number do you want a hard stop?
You can continue if…
- I can point at the counters and cost estimate in monitoring_runs.
- I know what budget_paused does and how to resume safely.
Done when: you can compute a run's cost from the formula, tell exact values from estimates, and know what to do for each typical failure.
Step 10
The final result check
Closure: a full review and the package you keep. From today you run the radar — Kydro remains the guide you return to.
Why this step matters
The last step teaches nothing new — it confirms what you have. Walk the checklist and complete the result card: a record of what you truly built and now manage.
What to prepare
- Steps 1–9 finished.
Tips
Growth options (deliberately beyond this guide): more competitors, extra categories, a second market segment. Expand only after two consecutive weekly reports prove accurate.
10.1
The full review
Walk the final checklist (resource below) unhurried — every item is something that sounded like jargon a month ago: credentials only in n8n, the baseline, dedup, budget_paused, recommendation decisions. All of it is yours now.
Finally complete the result card — it unlocks once every step is done and stays with you as the index of what you own.
A question for you
What is the first business decision you'll make differently thanks to your own radar's data?
Kydro Resources
You can continue if…
- The final checklist fully passed.
AI Market Radar — Complete Setup Package
An honest index of what you built step by step: accounts, a safely stored key, the workflows, the tables, your market configuration, the passed dedup test, alerts and the report — in one place. Kydro stores nothing for you; this summary stays with you.
- Accounts & the safe key
- Workflows & tables
- The dedup test
- Schedule & alerts
Done when: the final checklist passes in full and your setup package is complete.
Build your personal AI library.
Save useful goals and come back whenever you need them.
Success Tips
Expert recommendations to help you reach your goal faster and avoid common mistakes.
Start
RecommendedNarrow first, wider later
5 competitors and 4 categories give readable reports and low cost. Expand only after two accurate weekly reports.
Safety
RecommendedA spend limit before the first run
The OpenAI limit + the radar's action budget form a double net. Set both before you run anything.
Quality
Worth knowingRead the “Needs verification” section
That's where conflicts and undated findings live. Five weekly minutes there protect you from deciding on bad data.
Hygiene
Don't tune prompts for “more results”
The radar exists to filter, not to pad. Empty weeks are part of the market's truth.
Decisions
RecommendedClose the recommendation loop
Accept or reject every recommendation and fill the success metric — after a month you'll see which action types work for you.
Questions & answers
What people ask after finishing this goal
The questions that only come up once you have the whole picture.
No. Kydro teaches you to build your own monitoring — the automation runs on your n8n account, with your key, under your control. Kydro has no access to your n8n, your results or your data.
No. You import ready workflow files, paste prepared prompts and schemas, and every step says exactly where to click. Technical terms are explained at first use.
Yes — your own. That's where your radar lives. The guide uses n8n Cloud (Starter is enough); n8n costs are paid directly to n8n.
No. They are two separate products with separate bills. The radar needs an OpenAI API account with active billing and a spend limit — regardless of whether you have ChatGPT.
Only in your n8n: Credentials → OpenAI. Nowhere else — not in Kydro, not in a chat, not in a file, not in a form. If anything asks for the key, stop the process.
No. Kydro never collects or stores credentials, and the workflow files carry only the credential TYPE name — never a value. The key exists solely in your n8n Credentials store.
With 5 competitors and 4 categories a typical run is usually tens of cents (an estimate at 2026-08-01 prices; formula in step 9). The baseline run tends to be 1.5–2× pricier. The exact cost is always confirmed in the OpenAI billing panel.
Yes, twice over: the OpenAI spend limit (a hard monthly ceiling) and the radar's own search-action budget — once reached, the run enters budget_paused and nothing paid starts anymore.
A photo of the market from the first run: everything publicly out there at that moment. It is the reference point — only against it can later runs tell genuine news from things that had been online for ages.
Because that's correct. Run 1 builds the baseline — and a baseline is history, not news. An alert from the first run would mean the system confuses old with new.
No — it's a first-class, common result. The report then states plainly how many sources were checked and that nothing meaningful changed. Markets don't move every week.
Yes. It's best-effort monitoring of public, searchable sources — it cannot see closed groups, logged-in content or unindexed pages. Treat the radar as a strong complement to your market knowledge, not a replacement.
It can — which is why every recommendation is labelled an AI conclusion, carries a confidence level, evidence and a “needs verification” field, and execution always belongs to you. Nothing happens automatically.
The signal is marked “contradicted”, lands in the verification section with both versions and links, and the alert is blocked until you resolve it. A conflict always waits for a human.
No. The radar reads public sources only, stores no full articles or review authors' data, and is not a tool for gathering information about private individuals.
A running run: the Stop button (checkpoints keep progress; a resume finishes only what's missing). The whole schedule: switch the workflow to Inactive — a full, reversible stop. A total reset: additionally revoke the key in OpenAI.
The node retries. If the error persists, the task is marked failed with a reason and the run ends as partial — resumable safely, without duplicates. History loses nothing.
Yes, anytime: statuses in the competitors table (approved/archived) and the category list in the profile. A new competitor gets its own baseline on the next run — its history starts from zero.
Never. The radar doesn't contact competitors, publish content, change prices or message customers. Every recommendation is “proposed” and waits for your decision: accepted or rejected.
Dashboard and Access to Results
No. “Build Your Own AI Market Radar” provides the instructions, n8n workflows, prompts, data schemas and files needed to run the Market Radar AI engine. Once deployed correctly, the system can gather information, analyse signals and prepare reports, alerts and recommendations. A separate application with a graphical dashboard is not part of this guide.
The results appear wherever you configured them during the workflow setup — for example in a sheet, a database, a file, a message or a recurring report. n8n is mainly for running and controlling the automation. It does not replace a dedicated analytics dashboard.
No. The engine built with this guide can gather data, analyse it and deliver results without a separate dashboard. A dashboard is an optional layer that makes it more convenient to browse signals, competitors, recommendations, reports and the scan history.
A separate guide will be dedicated to exactly this: “Build a Dashboard for Your AI Agent”. It will walk you through creating the application, preparing the data storage, connecting it to the Market Radar engine and checking that the panel shows real results.
“Build a Dashboard for Your AI Agent” is an optional extension. It will not be required for the Market Radar AI engine to work.
Yes. It will be possible to prepare a local dashboard opened in the browser, for example at a localhost address. That variant can use a local JSON or CSV file, or an SQLite database.
A local dashboard is available primarily on the computer it runs on. The code, installation and configuration of that panel are not part of this guide.
The dashboard has to receive data from wherever the Market Radar workflow stores its results. You will need:
– an agreed data format; – a target storage location; – an adapter or an extra output stage in the workflow; – the dashboard connection configuration; – secure storage of the access credentials; – a test confirming that a record written by n8n appears in the panel.
The “Build a Dashboard for Your AI Agent” guide is meant to walk you through these actions step by step.
Yes, provided it is correctly connected to the Market Radar's data source. It can check for updates automatically, periodically, or after clicking a refresh button.
Merely starting a dashboard without a configured data source will not provide updates. The panel should always show the time of the last successful sync and a clear message when data is missing or the connection fails.
Yes, if the dashboard is made available online as a responsive PWA. You will be able to open it in a browser and add it to your phone's home screen without publishing a native app in the App Store or Google Play.
A local panel running only on your computer will not automatically be available on your phone. A mobile version requires online hosting, a secure connection to the data source and user authentication.
It depends on the configuration you choose:
– if n8n and the data source run locally, the computer must be on while scanning and saving results; – if the workflow runs in n8n Cloud and data is stored online, the automation can work with your computer switched off; – a local dashboard is available only while the computer and the app are running; – an online/PWA dashboard can be reachable from your computer and phone regardless of whether your own computer is on.
Not necessarily. A simple local variant can run without paid hosting, but it requires preparing, installing and maintaining the application on your computer.
An online/PWA variant may generate costs for hosting, a database, a domain, authentication or heavier service usage. Many tools offer free starter plans, but you should check their current limits and prices before deploying.
“Build Your Own AI Market Radar” gives you a working engine and an organised way of receiving its results. A graphical dashboard for your computer and phone is an optional next stage you will complete in the “Build a Dashboard for Your AI Agent” guide.