Sentiment Analysis Agent
Not just an LLM's opinion -- 14-source verification. Combines lexicon-based sentiment (VADER, AFINN), a transformer model (HuggingFace DistilBERT), toxicity (Google Perspective), entity and location verification (Wikidata, OpenStreetMap), news and community alignment (GDELT, Hacker News), grammar, readability, and language detection into one deterministic overall sentiment, urgency, and business impact score -- with a recommended action attached. Returns structured JSON ready for APIs, CRMs, ticketing systems, data lakes, and workflow automation -- callable synchronously or embedded inside asynchronous pipelines. Expect around 40 seconds per call given the real, parallel verification work.
Consume it however you already work
What a real result looks like
A real, live result -- fetching a fresh, current run now…
2 sources need a domain in the text (RDAP, PageSpeed); occasionally a real news-alignment lookup times out. Every result honestly reflects the specific text submitted -- these numbers vary call to call.
Honest, source by source -- not everything applies to every text
| Source | Type | Applies |
|---|---|---|
| VADER lexicon | Local | ✓ always |
| AFINN + Emoji Sentiment | Local | ✓ always |
| HuggingFace DistilBERT | Transformer | ✓ always |
| Google Perspective (toxicity) | API | ✓ always |
| Wikidata (entities) | API | when a name is detected |
| OpenStreetMap (locations) | API | when a place is detected |
| GDELT (news alignment) | API | ✓ always |
| Hacker News (community alignment) | API | ✓ always |
| LanguageTool (grammar) | API | ✓ always |
| Language detection | Local | ✓ always |
| Readability (Flesch) | Local | ✓ always |
| Lexical categories | Local | ✓ always |
| RDAP (domain trust) | API | when a domain is mentioned |
| Google PageSpeed | API | when a domain is mentioned |
The real, live dimensions_measured field always shows how many of the 14 genuinely succeeded for your specific text -- never inflated, never silently guessed.
No single model sees the whole picture
One model can hallucinate. One lexicon misses context. One news source misses global events. That's why ForceDream verifies across 14 independent systems -- statistical models, a transformer, lexicons, entity verification, knowledge graphs, a toxicity classifier, news signals, community discussions, grammar analysis, readability, and domain trust -- grounding the result in multiple independent sources rather than one AI prediction.
Real, independent sources -- not ForceDream's own opinion
Deterministic aggregation, not an opinion
Averages up to 3 independent, real signals -- VADER, AFINN, and the HuggingFace transformer -- into one normalized −100..+100 score. Confidence reflects how many of the 3 genuinely agreed on direction, not a fabricated certainty.
Built from negative sentiment strength, measured toxicity above 30%, and detected anger-related words -- a real synthesis of signals already gathered, not a separate, invented judgment.
Combines urgency with whether a real, verified brand/entity was mentioned and whether real, ongoing news coverage exists -- distinguishing an isolated complaint from a wider pattern.
Emotion and intent are the one genuinely subjective part -- LLM-derived, explicitly labeled as such, never presented as independently verified the way the rest of this page is. Every decision is traceable back to the individual sources that produced it. Individual verification failures never invalidate the whole analysis -- they are reported transparently and excluded from scoring.
What your money pays for
A verified pipeline vs. a single sentiment score
| ForceDream | Single-signal sentiment tools | |
|---|---|---|
| Sentiment score | ✓ | ✓ |
| Explains why the score exists | ✓ | × |
| Multiple independent signals cross-checked | ✓ | × |
| Toxicity scoring | ✓ | × |
| Entity/location verification | ✓ | × |
| News/community alignment | ✓ | × |
| Deterministic urgency + business priority | ✓ | × |
| Cryptographic proof | ✓ | × |
Nearly every plain sentiment API returns a bare number and nothing else. This one shows the reasoning behind it.
Built for more than a single support inbox
Also a genuine fit for enterprise CX teams, call centres, banks, insurance, healthcare, government, and trust & safety teams that need audit-ready, sourced sentiment rather than a single, unexplained number.
What support and CX leads actually care about
What you can analyze today
This agent analyzes existing text -- it doesn't generate or moderate content for you.
A real support-triage example
A real, direct API call you can drop into any workflow today -- no pre-built GitHub Action published yet.
14 sources run in parallel -- one slow or failing source never blocks the rest.
Individual verification failures never invalidate the whole analysis -- they are reported transparently and excluded from scoring.
Two lines, no custom wiring
Claude Desktop, Cursor, or any MCP client
Then call the real forcedream_generate_sentiment tool with text.
Real prompts you can adapt
Common questions
Standard A2A JSON-RPC
Real, live, ES256-signed
Every call settles atomically, with a real proof
Try it with your own key
Real call, real charge (£4.00, based on actual work done), real result -- may take up to 40s given 14-source verification plus LLM analysis. No ForceDream key? Sign up for a real trial balance.