Seba and Musa
They're aimed at different problems: Musa is a hosted platform for curated AI tours, and Seba is a lighter object guide that recognizes an exhibit from a photo and discusses it.
|
Musa Guide |
Seba |
|
|
Getting started |
Scan a QR code or reuse existing guide numbers. Audio starts at once, with no permission prompts. |
Open the web app, then allow the camera (and the mic for voice). |
|
Finding the exhibit |
A curated route with location-aware prompts, entered by QR code or number at each stop. Its web site doesn't mention photo recognition. |
The visitor photographs any object, and it's matched to your reference photos. No signage is needed. |
|
Experience |
A guided tour plus open questions, with context kept across the whole visit. |
An introduction and Q&A per exhibit. Context resets at each new exhibit, and there's no route. |
|
Content setup |
Upload catalogues, PDFs and floorplans. Musa builds a knowledge graph, and you set routes, personas and guardrails in its Studio tool. |
Staff add reference photos and a text description per exhibit on a simple admin page. |
|
Answer control |
A closed knowledge base with guardrails, citations and review before publishing. Open Q&A can be switched off. |
Answers come from the description. A venue setting limits or widens knowledge. There are no citations or review workflow. |
|
Languages and voice |
40+ languages from one source. |
4 (English, German, French, Spanish), using the phone's built-in voices. More can be added. Voice input works only on Chrome for Android and Safari on iPhone. |
|
Analytics |
Dwell time, tour completion, top questions by gallery, revenue. |
Dwell time, tour path, Identify failure rate. |
|
Offline |
Tour audio plays offline. Live answers pause without a connection. |
Needs a connection. |
|
Business side |
Revenue share, £500/month or prepaid from £1,000, with built-in payments and ticketing/CRM links. |
Not productized, and it serves a single venue. |
|
Track record |
Live at Museo Miraflores, where it launched in eight weeks with roughly 5x higher uptake than handheld devices and sessions over 40 minutes, per Musa's internal data. It also claims a 2025 blooloop award. |
Proof of concept |
· Proven in service: It's a live, paid deployment with staff-facing tools for analytics and payments.
· Answer trust: Guardrails and citations, plus curated storytelling, are a strong answer to "can we trust the AI?"
· Low visitor friction: There's nothing to install and no permission prompts.
· Mostly self-reported: The results and award come from Musa itself, and I found no independent test of answer accuracy or voice quality.
· Setup effort: It needs QR codes or numbers at every stop plus curated content, over a few weeks of onboarding.
· Ongoing cost and dependence: It's a rented platform. It offers export, but you don't control the code.
· Framing: Musa's marketing calls object-level chatbots "one painting at a time", which is Seba's category, so expect that critique.
· Photo recognition: It works on any object with no QR codes or labels, and staff can add a new exhibit in minutes.
· Personalization: Visitors set age group, knowledge level, response length and language.
· Staff tools: They include an Identify test tab with feedback, plus duplicate-photo warnings.
· Ownership and cost: You own the code and data, and running costs are low.
· Identification is the weakest link: Look-alike exhibits, glare and bad lighting can fool it. Your own test log showed a confident wrong match, so each exhibit needs several varied photos.
· No tour structure: There's no route or "what next" guidance, and context resets per exhibit.
· No offline mode and no analytics dashboard.
· Narrow reach: It has 4 languages and depends on the phone's built-in voices.
· Thin content controls: There are no guardrails, citations or review step, and nothing that imports catalogues or PDFs.
· Single venue: One shared admin key protects everything.
· Unproven at scale.