Setting up Seba for a new Gallery
A short guide for museum directors and curatorial staff on using Seba for a new gallery.
Everything Seba says comes from a handful of plain text files placed in the app's content folder there is no coding involved. These are exactly the kind of text that curatorial staff writes, reviews, and updates regularly. They are the single biggest factor in a good visitor experience.
· venueTheme.txt an overview of the gallery or collection as a whole, roughly 200300 words. This is the general context Seba keeps in mind for every conversation, regardless of which exhibit a visitor is looking at the institution's character, the scope and strengths of the collection, and any historical background worth knowing. Written once per venue, and rarely needs to change.
· exhibit data one plain text file per exhibit (named exhibit001.txt, exhibitGuanyin.txt, or similar any unique name starting with exhibit):
· Everything is free-form Seba reads it directly, so plain, unformatted prose is completely fine. No headings, bullet points, or layout needed. A good file typically includes the basics (title, date, culture or period, artist or maker, medium, dimensions, provenance) and then goes deeper: history, significance, symbolism, stories worth telling, restoration, controversies, and cultural context.
· Length: 3001000 words is a good range, and more detail is generally better Seba automatically adjusts its own vocabulary and depth to match each visitor's stated age and knowledge level, so richer source material simply gives it more to draw on for visitors who want to go deeper, without changing what a casual visitor hears.
· Mention related exhibits by name nearby pieces, similar works elsewhere in the collection, pieces from the same period or maker. This lets Seba draw comparisons and point visitors toward other things worth seeing. It's worth cross-referencing in both directions: mention exhibit A in exhibit B's file, and B in A's.
· What to avoid: don't state uncertain or disputed facts as settled Seba is instructed to say plainly when something isn't documented rather than guess, so it's better for a file to leave a gap honest than to fill it with speculation. Likewise, don't include anything the museum wouldn't want spoken aloud to a visitor of any age or background, since Seba may reference or repeat this material if asked.
· Language The description text can be in English, German, French or Spanish. Regardless, Seba will translate and speak to the visitor in their preferred language.
· Adding more exhibits no code changes are needed. Simply add another exhibit file to the content folder and Seba picks it up automatically. Removing an exhibit works the same way in reverse just delete or move its file out of the content folder, no code changes needed there either. One approach worth considering: start by having Seba cover a handful of exhibits and gradually add more over time. If taking that route, mark or tag the exhibits currently covered (a small sticker or icon near the label works well) so a visitor doesn't scan a piece that isn't set up yet.
· Testing the material before opening a new gallery or adding a batch of exhibits, walk the floor and scan each exhibit's actual label, confirming Seba identifies the correct piece every time. Pay particular attention to keyword collisions two exhibits with similar names or overlapping descriptive language can be confused for one another, especially as more exhibits are added over time. If that happens, make the KEYWORDS line more specific to that exact item's label wording rather than a generic term shared with others.
· venueOptions.txt an optional file for venue-wide settings. Each setting is written as KEY = value. Lines starting with # are treated as comments and ignored, so the file can safely include your own notes.
· KNOWLEDGE = LIMITED, or GLOBAL LIMITED (default) restricts responses to provided exhibit descriptions. GLOBAL allows responses to includes generally known information. Use with care!!
· DURATION = 15, 25, or 40 sets the default length, in seconds, of Seba's spoken answers. Visitors can still adjust this themselves in the app; this just sets the starting point.
· SILENT = FORCE locks Silent Mode on for every visitor and is useful for galleries where quiet is expected or enforced. This mode hides the voice button, and removes the ability for visitors to verbally ask questions. Visitors can still choose the preselected questions or type in their questions. The response will be spoken and should be heard using headphones.
· venueHeader.txt the very first thing a visitor sees when they open the app, about 1025 words. It can be in any language. A brief welcome plus a reminder to scan an exhibit label to get the visitor started.
Seba runs in a web browser, so most visitors can simply use their own phone no app to install. That said, a few practical choices make a real difference to the experience on the floor.
Most phones and browsers are supported. A full breakdown, including how a visitor grants camera, microphone, and sound access on their own device, is covered in the companion Visitor Guide, which can be handed to visitors directly. In general, Android and Chrome running on a modern phone presents the best user experience. Apple devices work well, but require additional button presses.
Headphones are worth encouraging even a simple, low-cost pair makes a noticeable difference. They keep the gallery quiet by avoiding a room full of overlapping phone speakers, they give visitors clearer audio in echoey, high-ceilinged, or busy spaces, and they make the experience feel more personal closer to a private docent than a public announcement. Headphones also pair naturally with Silent Mode: a visitor can listen privately even in a gallery where speaking aloud isn't appropriate, asking follow-up questions by tapping or typing instead.
Consider lending or renting tablets and headphones for the smoothest experience across every feature (camera scanning, spoken introductions, and spoken follow-up questions), Android phones running Chrome are recommended. If budget allows, a small fleet of low-cost Android devices, pre-configured with camera and microphone permissions already granted, removes any setup friction for visitors who'd rather not use their own phone, or whose phone or browser doesn't support every feature. Worth keeping in mind: older, lower-powered devices tend to run slower and can degrade the visitor's experience, so it's worth choosing reasonably current hardware even if it doesn't need to be top-of-line.
The software for Seba is free. However, Seba is currently powered by Anthropic's Claude AI which incurs a small, usage-based cost each time Seba generates a spoken response. To put a number on it: consider an average visitor who views 15 exhibits and asks 3 follow-up questions per exhibit. Counting the automatic spoken introduction Seba gives for each exhibit, that's 4 AI responses per exhibit or 60 responses across the visit. Based on real, measured usage during development, each Claude response costs roughly $0.0069 on average. For that visitor, a full visit comes to approximately $0.42. Scaled up, that's roughly $42 per 100 visitors per day or about $15,000 a year for a gallery seeing 100 such visitors daily. The $0.0069 figure comes from limited testing rather than a large-scale deployment. Real-world costs may be lower, since responses to popular exhibits will get reused efficiently across many different visitors' conversations throughout the day.
There is always a price to performance tradeoff: there are several free AI services available that have not yet been tested. The reliability, restrictions and response quality of these services need be checked. Lastly, the speech recognition function can be greatly improved, at a cost, with Wispr Flow.
One more thing worth considering, once Seba is running well on the gallery floor: the same approach could extend naturally to the museum's public website.
Many museum websites fall into one of two unsatisfying extremes for exhibit information: a single thin paragraph that leaves a curious visitor wanting more, or a dense scholarly PDF that only a specialist would read start to finish. Either way, the page is static it says the same thing to every visitor, and there's no way to ask it the one question you actually have.
Imagine instead a visitor browsing the collection from home, touching a photo of an exhibit on their screen, and simply starting a conversation about it the same guide, the same adaptive depth for a curious teenager or a specialist, available to anyone researching the collection at any hour, not only to people standing in the gallery.
The appealing part is that this wouldn't mean starting over: the exhibit content built for Section 1 the venue overview, the per-exhibit detail, the connections between related pieces is exactly what would power this too. The same investment would do double duty.