Tisane Labs

@tisane analyze this customer review for sentiment and key topics: 'The product works fine but shipping took forever and support never responded to my emails.'

This example would return a structured NLU breakdown: overall sentiment detected as mixed (positive on product quality, negative on logistics and support), entities extracted (product, shipping, support, emails), and topics identified (customer service issues, delivery delays). The response might flag 'never responded' as a complaint signal and 'works fine' as a positive attribute, giving you actionable insight into what's driving the review's tone.

This showcases Tisane's core strength—multi-dimensional text analysis beyond simple positive/negative scoring. Requires an API key with sufficient quota; complex texts may consume more credits. The output structure is rich (sentiment, entities, topics, abuse detection) so you'll want to specify which dimensions matter most for your use case to avoid information overload.

@tisane translate this product description to French and then paraphrase the French version to sound more casual for social media.

This example would first transform the English description into formal French, then apply a paraphrasing pass to adjust register and tone—shortening sentences, using colloquial phrasing, and adapting idioms. The output might show both the literal translation and the paraphrased variant side-by-side, illustrating how the same content can be reframed for different audiences while preserving core meaning.

Demonstrates Tisane's dual capability: cross-language translation plus same-language style transformation. The paraphrasing quality depends on the target language's support level (call get_supported_languages first). Be aware that casual paraphrasing may lose precision—review output before publishing, especially for regulated content like medical or legal copy.

@tisane I have 50 support tickets in HTML format. Extract the plain text from each, detect the language, then calculate semantic similarity between tickets to find duplicates.

This example would chain three Tisane tools: extract_text strips HTML tags and decodes entities to get clean prose; detect_language identifies which tickets are in English, Spanish, etc.; calculate_semantic_similarity runs pairwise comparisons (or compares each ticket to a reference) and returns similarity scores (0.0 to 1.0). Tickets scoring above 0.85 might be flagged as likely duplicates, letting you merge or prioritize responses efficiently.

This workflow highlights how Tisane's tools compose—cleaning messy input, routing by language, then finding semantic overlap. With 50 tickets, you're running 50 extractions + 50 detections + potentially 1,225 pairwise comparisons (50×49/2), so API rate limits and token costs matter. Consider batching or sampling if your ticket volume is higher, and confirm your API tier supports the call volume before running at scale.

When Tisane Labs beats keyword filters for support triage

A 6-person SaaS support team gets 200 tickets daily across email, chat, and in-app forms. Half arrive in Spanish or French. They need automatic routing to billing, technical, or sales queues before a human sees them. Tisane Labs wins here because the Analyze Text tool extracts sentiment and topics in one call, and Detect Language handles the multilingual mix without pre-filtering. The team sets up a workflow: incoming ticket → detect language → analyze for entities and topics → route by topic tags. The threshold: if your ticket volume is under 50/day, a simpler keyword tagger is cheaper. Above that, Tisane's NLU accuracy pays off in fewer mis-routed tickets. Connect it to your helpdesk API and let the MCP handle the linguistic heavy lifting.

Using Tisane for real-time comment filtering at scale

A 12-person team runs a niche social platform with 8,000 daily user comments. They need to flag abusive language, spam, and off-topic posts before they go live. Tisane Labs fits because the Analyze Text tool returns sentiment, entities, and linguistic features in a single pass—no chaining multiple APIs. The workflow: user submits comment → Tisane analyzes → flag if sentiment is strongly negative or specific entity types appear → queue for human review or auto-hide. The trade-off: Tisane's 6-tool scope is overkill if you only need sentiment scores; a dedicated sentiment API is faster. But if you're also extracting topics or handling multiple languages, the unified NLU model saves integration time. Best for platforms where context matters more than raw speed.

When Tisane's similarity scoring catches translation drift

A 4-person product team ships docs in English, German, and Japanese. After each release, they need to verify that translated strings still match the source meaning—especially for legal disclaimers and feature descriptions. Tisane Labs works here because the Calculate Semantic Similarity tool gives a 0-1 score comparing source and target text, and Transform Text can paraphrase edge cases for human review. The workflow: export doc strings → run similarity checks on English vs. translations → flag scores under 0.85 → review flagged pairs. The boundary: if your docs are under 500 strings per language, manual spot-checks are faster. Above that, Tisane's batch processing through the MCP saves hours per release cycle. Pair it with your CMS or translation memory tool for automated QA gates.