The alternative to Talkwalker or Brandwatch that doesn't cost thousands a month. 21 data sources, an AI engine of your choice, and a set of signals enterprise tools don't have at all — self-hosted, bring-your-own-keys, running in 30 seconds flat.
Bilingual (English / Italian), on two independent axes. One flag switches the interface; a separate one sets the language the AI writes in — briefs, Point of View, answers, narratives. They are deliberately decoupled: you can read Radar in Italian and still generate deliverables in English, because the language of the output depends on who will read it, not on who is using the tool. Your collected data is never translated: mentions, topics and sentiment stay in their original language, because the interface language must not alter the dataset. More languages are welcome as contributions — a missing translation simply falls back to English.
The whole conversation as a real 3D solar system — sources are planets, sentiment split becomes moons, the sun is your Health Index:
Source → Topic → Sentiment flow, with multi-select cross-filtering:
Real-world geographic map and the executive Health Index:
Pick your AI engine — Claude, OpenAI or Grok — enter its key and set the models, all from the app. No code, no redeploy:
Point of View — 90 days of data turned into a thesis you can defend, in slide-ready blocks with clickable citations, plus the Market × Research crossover that shows where academic research is running ahead of the market:
Earned Media Value — what your coverage would have cost to buy, explained in plain language and with every assumption on screen:
The insights hub: 16 visualizations grouped by the question they answer, so the sidebar stays readable and every chart says what it is for:
Zoom into any single channel with a source deep-dive — how that channel's volume, sentiment, topics and authors compare with the whole project, each figure a verifiable subset of the totals:
A Radar project starts from one of two places, and everything downstream — charts, insights, reports, exports — works the same either way.
Listening: Radar collects the conversation itself from public sources, and you read it as it arrives — relevance stars, sentiment, topics, translation on demand:
Import: you already have the data. Drop in one Excel or CSV export — or twenty, with
every sheet shaped differently — and Radar reads what each column actually contains
instead of demanding a template. Here one workbook has become four sheets: two of posts,
one of monthly follower counts, one of personal-branding metrics, each with its own
mapping. Cells produced by a formula are stored as their result, and the columns
Radar has no field for (PILLAR, CAMPAGNA, editorial format) are kept as custom fields
rather than dropped:
Counts alone don't answer that — you have to take them on faith. So Radar does the check you would do yourself: it takes the first row of your sheet, one in the middle and the last, and puts your file next to what is actually in the archive, field by field. It is not a simulation — it goes and fetches the stored record back:
The headline numbers count what is in the archive right now, not what the import log once claimed to have written. When the two diverge — a sheet imported before you changed a column assignment — the panel says so in plain language and offers the one gesture that fixes it: re-import. The original file is still there; nothing needs re-uploading.
When a sheet is about people rather than channels, Radar recognises it and builds the personal-branding view: the team at a glance, audience growth per person, and a focus on any single one. (Fictional names and data.)
Every other view in Radar answers a question we chose. Studio Graph lets you ask your own: pick the X, Y and Z axes from the fields your project actually has — listening fields and spreadsheet columns in one list, because it doesn't matter to the reader where a number came from — then pick the shape, the aggregation and the palette.
Three things it does deliberately:
- No SQL ever leaves the client. You choose from a closed catalogue of fields; the server translates each id into an expression. A custom column name travels as a parameter, never as concatenated SQL.
- It warns, it doesn't forbid. Summing a rate, a pie with twelve slices, a line drawn between categories that have no order, more than eight series — each is stated in plain words under the title, and you decide. Sometimes the unusual choice is the right one.
- The palettes are the three jobs of colour, not three moods: identity (categorical), intensity (sequential), polarity (diverging). All three pass a colour-vision-deficiency check — and the categorical one is validated as an ordered sequence, which is why there is no "reorder" option: the same eight hues in a different order put pink next to green at ΔE 1.6 for a deuteranope, i.e. the same colour.
Saved charts show up in Explore insights alongside Radar's own — with an (X) to remove them, which the built-in ones don't have, so the difference needs no explaining:
Compose a report page by page: pick charts from the catalogue — Radar's own and the ones you built in Studio Graph — reorder them, and let the AI write the commentary. Each comment has a role you choose before generating it: a presentation before the chart, a comment after it, both, or a synthesis of how the charts on the page relate.
The model never sees your rows. It receives the figures already computed by SQL and writes prose around them — the same rule as the Point of View. A report is a running order, not a snapshot: re-export it in a month and it redraws with that month's data.
The same charts reach the generic export too — PDF, Excel, Word and PowerPoint, each rendering what suits it (a table in Word, a sheet of long-format rows in Excel, a bar slide in PowerPoint):
Every exported file carries the AI Act disclosure (Reg. (EU) 2024/1689, art. 50) in its notes and in its file metadata — not in the title and not in the body, where it would read as part of the analysis.
Not everything worth watching is a keyword match in a stream of posts. Three self-contained sections, each with its own sources and its own kind of analysis, answer questions that a pure mention-search tool structurally can't:
Sport — for publicly listed clubs, match results crossed with fan sentiment and share price on one timeline. Three incomparable units (league points, sentiment, euros) on hidden axes, so the only question that matters is whether the lines move together — plus how long a win's mood boost actually lasts, and whether the margin of victory matters at all:
Wikipedia edit monitoring — who is editing a brand's Wikipedia page, and does it look like a fight? Reverts, anonymous accounts and a weekly activity chart surface an edit war before it shows up anywhere else — no AI, just the official MediaWiki API reading its own revert/anonymous tags:
Reviews — star ratings from the App Store, Google Places and Yelp in one place, because a rating is the sentiment already, no AI needed to interpret it:
Share of search — Google Trends interest for your brand against its competitors, already scaled against each other (100 = the peak across all of them), sitting next to Share of Voice on the Benchmark page:
Enterprise listening tools cost thousands per month. Radar gives a single person or a small team the same core workflow — listen, analyze, decide, create — using public data sources and the Claude API, for the price of an API key (or nothing at all if you just want to collect data).
- Your AI, your choice: run every AI feature on Claude (Anthropic), OpenAI or Grok (xAI). Pick the engine, paste its key, and even set which models power the fast (bulk tagging) and smart (briefs, insights) tiers — all from Settings → Budget, no code and no redeploy. Switch providers anytime; the spend cap works the same for all of them.
- 21 connectors, 15 free (11 need no key at all): worldwide news via GDELT (100k+ outlets, 65+ languages) and Google News, Bluesky, Mastodon, Hacker News, Telegram, RSS, Stack Exchange, GitHub, SEC EDGAR filings and arXiv papers out of the box — plus Reddit, YouTube, Discord and public LinkedIn posts with a free API key. Full table below. Every source is verified live against its real API before shipping, not assumed from documentation — see CONTRIBUTING.md for what that means in practice.
- Public LinkedIn listening, legally: LinkedIn offers no public post search, so Radar finds public LinkedIn posts and articles by anyone through the Tavily search index (free tier, 1,000 searches/month) — labelled LinkedIn (web) and kept distinct from LinkedIn (page) (your own company page via the official API), so the two acquisition models never get mixed.
- 6 premium connectors ready (X, Instagram, Facebook, TikTok, NewsAPI, LinkedIn page) — drop in a key and the source turns on.
- Beyond mentions ✦: three self-contained sections outside the keyword-search model — Sport (match results × fan sentiment × share price for listed clubs), Wikipedia edit monitoring (edit wars and anonymous activity as an early-warning signal) and Reviews (App Store, Google Places, Yelp star ratings). See the screenshots above.
- Share of search: Google Trends interest for every benchmark entity, already scaled against each other, next to Share of Voice — because how much people search for you matters as much as how much they talk about you.
- Import your own data: create an Import project and drop in an Excel/CSV export (from any platform or vendor). Radar runs the whole analysis engine — sentiment, topics, narratives, insights — over rows it never scraped, so external data lives alongside live listening. Many files, many sheets, each shaped differently: Radar reads what the columns contain rather than requiring a template, keeps formula cells as their result, preserves the columns it has no field for as custom dimensions, and can hand you the normalised file back. Nothing is consumed — the raw rows stay, so a re-mapping is a re-import, not a re-upload. See the walkthrough above.
- A second data type: measures — not everything in a spreadsheet is a post. Aggregate sheets (monthly followers, publishing mix, engagement rate per manager) become who · what · when · how much, wide or long, and get their own charts. Rates are averaged, cumulative totals take the latest value, counts sum — because summing a rate produces a number with no meaning.
- Studio Graph ✦: build the chart nobody planned for — choose the X, Y and Z axes from your project's real fields (listening and spreadsheet columns in one list), the shape, the aggregation and the palette. It warns instead of forbidding, never lets SQL through from the client, and saved charts turn up in the insights hub, in the custom report and in all four export formats. Screenshots above.
- Point of View ✦: turns 90 days of data into a defensible thesis for a meeting — 3-5 slide-ready blocks (named idea, narrative, supporting figures), counter-signals and clickable citations to the real posts. The numbers come from SQL, never from the model, and invented citations are dropped by a validation pass. It also crosses your market signal with academic research (via the open OpenAlex index) to show where research is running ahead of the market, and where the market is loud with no research behind it.
- Message pull-through: whether the messages you want to land are actually picked up — by which sources, with what tone, and where exactly they landed.
- Earned Media Value: what the coverage would have cost to buy, computed conservatively with every assumption on screen (negative coverage valued at zero and reported separately, no arbitrary AVE multiplier).
- Per-source deep-dive: focus on one channel and compare its volume, sentiment, topics and authors against the whole project — every number a verifiable subset of the totals, with a banner spelling out how that source's data is collected.
- AI analysis with Claude: sentiment, emotion (joy/trust/fear/anger/sadness/ surprise), relevance scoring, story clustering, daily executive briefs, conversation clusters, cause-effect chains, quality scores.
- Brand vs Market Health Index: one 0–100 score for the whole theme, and — when you flag one benchmark entity as your brand — your brand's score next to the market and a competitive ranking.
- 12 advanced insight visualizations (see below), from a live 3D "Conversation Galaxy" to a Share-of-Voice streamgraph, a Source→Topic→Sentiment Sankey and a real-world map.
- Content Studio: turn a concept into a multi-format kit, explore alternative hooks, refine drafts conversationally in your brand voice.
- Cost control: an admin-set spend cap on the AI, a password-protected reset, and an all-time total across every user — the app stops calling the AI at the cap while data collection keeps running.
- Custom report with AI commentary ✦: compose pages from the chart catalogue and your own Studio Graph charts, then have the AI write the text — choosing its role before generating: presentation before the chart, comment after, both, or a synthesis of how the charts on a page relate. The model only ever sees figures already computed by SQL. There is also a periodic edition (7 cadences) that freezes the Point of View it used and always states when that POV was written and on what data.
- Exports: branded PDF, PowerPoint, Word, Excel — every insight included, plus the charts you built yourself. Read-only share links. Each file carries the AI Act disclosure (Reg. (EU) 2024/1689, art. 50) in its notes and metadata.
No database, no keys, no cloud account required. Uses an embedded database (PGlite) and runs entirely offline. AI features stay idle until you add a key for your chosen engine (Claude, OpenAI or Grok).
git clone https://github.com/Scognamiglio1969/radar-intelligence.git
cd radar-intelligence
npm install
npm run dev
# open http://localhost:3000 · first login: [email protected] / changemeTo enable AI, add your own key to .env.local (or paste it later from Settings → Budget):
cp .env.example .env.local
# Claude: ANTHROPIC_API_KEY=... (https://console.anthropic.com)
# OpenAI: OPENAI_API_KEY=... (https://platform.openai.com)
# Grok: XAI_API_KEY=... (https://console.x.ai)One-click deploy to Vercel (add a free Neon Postgres and your keys when prompted):
Minimum production env vars: DATABASE_URL, SESSION_SECRET. Recommended:
ANTHROPIC_API_KEY, ADMIN_EMAIL, ADMIN_PASSWORD. See .env.example
for the full, commented list.
Radar never ships with anyone's keys — you bring your own, and everything is stored encrypted at rest.
- AI engine key — powers all AI features (sentiment, briefs, ratings, clustering,
Content Studio, "Ask the data"…). In Settings → Budget → AI engine choose your provider —
Claude, OpenAI or Grok — and paste its key, or set the matching environment variable
(
ANTHROPIC_API_KEY,OPENAI_API_KEYorXAI_API_KEY). The default models per provider are editable, so a newly released model just needs its id typed in — no code change. No key needed just to collect data. - Data-source keys (X, Instagram, Facebook, TikTok, LinkedIn, NewsAPI, Reddit, YouTube, Discord, Tavily for LinkedIn-web, football-data.org and Alpha Vantage for Sport, Yelp for Reviews) — configured from the UI in Settings → Sources (or inline on the Sport / Reviews pages, next to the source they unlock), or via environment variables as a fallback.
Nothing requires editing code or redeploying: an admin adds the keys from the app and the features turn on immediately.
The sidebar is organised by what you are trying to do — Monitor (what is happening), Analyze (understand the numbers), Interpret (what it means), Create (what you produce) and Setup — instead of by chart type. The advanced visualizations live in a hub that groups them by the question they answer, so the menu stays readable.
| Page | What it does |
|---|---|
| Dashboard | KPIs, volume per source, sentiment, emerging topics, latest brief |
| Listening | Stream of every mention with filters (source, sentiment, language, period, text) — filter by a source to open its deep-dive, or land on a curated set of posts from a narrative |
| Source deep-dive | One channel vs the whole project: volume, sentiment, topics and top authors, each a verifiable subset of the totals |
| Import | Create a project that ingests Excel/CSV files instead of scraping — many files, many sheets, each shaped differently. Guided step by step, with a spot check that puts your rows next to what is in the archive |
| Measures | The aggregate series a spreadsheet brings — follower growth, publishing mix, engagement rate — as who · what · when · how much |
| People | Personal branding: the team at a glance, audience growth per person and a focus on any single one, built from the sheets that are about people |
| Studio Graph | Build your own chart: choose the X, Y and Z axes from your project's real fields, the shape and the palette. Saved charts appear in the insights hub, the custom report and every export |
| Media | News grouped into stories (AI clustering) + most active outlets |
| Benchmark | Share of voice, trends and comparative sentiment across configurable entities |
| Audience | Most active communities, languages, influential authors, topics by community |
| Content | Engagement ranking (per-platform percentile) + AI quality score |
| Point of View | An evidence-backed market thesis in slide-ready blocks, with research corroboration |
| Message pull-through | Are your key messages being picked up, and by whom |
| Media value | Earned Media Value with every assumption exposed |
| Explore insights | A hub grouping 16 visualizations by the question they answer — plus the charts you built in Studio Graph, the only ones that can be removed |
| Content Studio | Concept → multi-format kit, Hook Lab, conversational refinement |
| Alerts / Brief | Auto-detected volume spikes & sentiment drops; daily executive brief |
| Custom report | Compose pages from charts and commentary — written by you or generated by the AI in the role you choose (before the chart, after it, both, or a synthesis of the page). Also in a periodic edition across 7 cadences |
| War Room | Full-screen live view for a wall display |
| Settings | Tabbed: My account, Team (mark one entity as your brand), Sources (connector status & keys), Budget (choose the AI engine — Claude/OpenAI/Grok — its key & models, spend cap, admin cost controls), Credits & Legal |
| Reviews (beyond mentions) | Star ratings from App Store, Google Places and Yelp, plus your own imported file — a self-contained section, since a rating already is the sentiment |
| Sport (beyond mentions) | For publicly listed clubs: match results crossed with fan sentiment and share price, an edit-anatomy of how long a win's mood boost lasts, and whether the margin of victory matters |
| Wikipedia (beyond mentions) | Who edits a followed page, how often, and whether it looks like an edit war — reverts, anonymous accounts, a weekly activity chart |
A suite of purpose-built visualizations, each on its own page and each included in the PDF / PowerPoint / Word / Excel exports:
| Insight | What it shows |
|---|---|
| Conversation Galaxy ✦ | The whole conversation as a real WebGL solar system: the sun is your Health Index, planets are sources (size = volume) with photographic NASA-based textures, and each planet has three moons sized 1–10 by its sentiment split. Drag to orbit, scroll to fly closer. |
| Health Index | One 0–100 composite (sentiment, positive share, momentum, resonance). Market health always; Brand health + competitive ranking when you flag your brand. |
| Share of Voice | 100% stacked area of each entity's share of the conversation over time, plus an explicit 30-day ranking. |
| Conversation flow | Source → Topic → Sentiment Sankey. Multi-select cross-filtering: pick any sources × sentiments to isolate a sub-flow (e.g. what drives negativity in two sources). Rich hover tooltips. |
| Momentum quadrant | Topics by volume × acceleration: Rising stars / Emerging / Steady / Declining. |
| Emotion radar | The emotional fingerprint beyond sentiment (joy, trust, fear, anger, sadness, surprise). |
| Geographic map | Real-world choropleth: each language shades its countries (English → US/UK/CA/AU, Spanish → Latin America…), colored by sentiment. |
| Semantic constellation | Key terms as a star map — size = frequency, color = sentiment, links = co-occurrence. |
| Influencer network | Force-directed graph of the top authors, clustered by community, sized by engagement. |
| Author pyramid | Authors tiered by influence (mega / macro / micro / long tail) with the share of reach each tier holds — is your conversation carried by a few big voices or broadly distributed? |
| Crisis radar | A risk gauge plus the anatomy of the biggest spike: what drove it, and the content that weighed most. |
| Topics · Heatmap · Waterfall · Clusters · Cause-effect | Topic×sentiment map, hourly heatmap, sentiment waterfall, conversation clusters, AI cause-effect chains. |
Every connector is a small, isolated file (lib/connectors/) — see
CONTRIBUTING.md for how to add one. free works immediately;
freekey needs a free key to turn on; premium needs a paid API.
| Source | Tier | Category | Notes |
|---|---|---|---|
| Google News | free | General & news | Public articles by keyword |
| GDELT | free | General & news | 100k+ outlets, 65+ languages |
| RSS | free | General & news | Any feed you add, unlimited |
| NewsAPI | premium | General & news | ~150k outlets, boolean search |
| SEC EDGAR | free | Financial & corporate | US-listed company filings (8-K/10-K/10-Q/DEF 14A) |
| arXiv | free | Academic | Papers with full abstract, re-checked for literal matches |
| Stack Exchange | free | Tech & developer | Stack Overflow + 5 other communities |
| GitHub | free | Tech & developer | Issues/PRs with full text, bot-filtered |
| Hacker News | free | Tech & developer | Stories and comments |
| freekey | Social & messaging | Posts and comments | |
| Bluesky | free | Social & messaging | Public posts |
| Mastodon | free | Social & messaging | Public toots |
| Telegram | free | Social & messaging | Your own watchlist of public channels |
| Discord | freekey | Social & messaging | Your own server, via your bot |
| LinkedIn (web) | freekey | Social & messaging | Any public post/article, via Tavily search |
| LinkedIn (page) | premium | Social & messaging | Your own company page, official API |
| X (Twitter) | premium | Social & messaging | Real-time search |
| premium | Social & messaging | Hashtag search | |
| premium | Social & messaging | Your own linked pages | |
| TikTok | premium | Social & messaging | Research API |
| YouTube | freekey | Video | Videos by keyword |
Plus, outside the mention model entirely: App Store, Google Places and Yelp (review ratings), football-data.org and Alpha Vantage (Sport), Google Trends (Share of search) and the MediaWiki API (Wikipedia edit monitoring) — see Beyond mentions above.
- Next.js 16 (App Router) — deploys on Vercel, dark theme
- Postgres: Neon in production, embedded PGlite for local dev (zero setup)
- Drizzle ORM, schema created & migrated automatically on boot
- Pluggable AI engine — Claude (
@anthropic-ai/sdk), OpenAI or Grok (OpenAI-compatible API), switchable from the UI, all behind one hard monthly spend cap - Recharts + hand-built SVG/Canvas charts, and three.js for the 3D galaxy
- Pluggable connectors (
lib/connectors/) — adding a source is one small file - Beyond mentions sections (
lib/sport/,lib/reviews/,lib/wikipedia/,lib/search-interest/) follow the same self-contained pattern as connectors — their own tables, their own ingest, their own page — for data that isn't a keyword match in a post
See CONTRIBUTING.md for a walk-through of adding a connector.
Planet/sun/moon textures in the Conversation Galaxy are © Solar System Scope, licensed CC BY 4.0.
Radar collects publicly available data. You are responsible for complying with the Terms of Service of each data source (Reddit, X, Meta, GDELT, Google News, etc.) and with applicable data-protection laws (e.g. GDPR) in your jurisdiction. Some sources restrict automated access; enable only what you are entitled to use. Note that Share of search uses Google Trends' internal, undocumented API (no official public API exists) — it is best-effort and checked at most once a day, not a guaranteed-uptime data source. This project is provided as-is, without warranty. See SECURITY.md to report vulnerabilities.
Issues and pull requests are welcome — see CONTRIBUTING.md and our Code of Conduct. Good places to start: internationalization, new connectors, tests. See CHANGELOG.md for how the project got here.
GNU AGPL-3.0 © Massimo Scognamiglio and contributors. If you run a modified version as a network service, you must make your source available under the same license.