For Museums & Historical Institutions

Let your collection speak for itself.

Historacle brings the people in your collection into the room with your visitor. They sit across from the figure and hear it in its own voice, grounded only in your sources, with a citation on every claim.

Start a conversationSee how it works

01The opportunity

In most exhibitions studied, the average visitor spends under twenty minutes (Serrell, 1997) and walks past the archive that holds the deepest stories. Your scholarship lives in PDFs no one opens. The opportunity is not to compete with the artifact. It is to give visitors the conversation that unlocks it.

02How it works

A deployment is not a chatbot on a page. Visitors sit across from the people in your collection, in a room rendered around them, and the figure replies in its own voice.

01

Visitors sit across from the figure and talk, by voice or by typing. It answers in its own register, only from documents you have approved.

02

Walk visitors into the room where a decision was made, built from a real artifact. The figure narrates; the moment is rendered around the visitor.

03

Sit in on a live debate across your collection, figures, schools, and eras, and choose when to interject a controversy your archive already documents.

04

Turn any conversation into flashcards, quizzes, short videos, and storybooks. Multi-sensory where your archive supports it, always sourced.

03The hard part, done right

Generative AI fabricates, even when it reads from source documents. So a Historacle figure only draws from what you have approved, ships the source with every claim, and refuses what the record does not support. When you cannot audit an AI, you should not ship it. With Historacle, you can.

  • Grounded in your archive only. No outside data, no general-web bleed.
  • A citation on every claim. Visitor and curator both see the source.
  • You author voice, scope, and refusals. The figure is yours, not a template.
  • Sensitive material handled deliberately. Contested histories and restricted content are gated during authoring, never improvised.
  • Your archive stays yours. No cross-tenant training, no third-party fine-tuning on your corpus, full export on exit.

04Deployment

A figure runs in a browser on any device, hosted end to end on our servers. It adds no load to your own site. A few of the ways institutions put their figures in front of visitors today:

Your own website

Your URL, your branding. Institutions embed their figure on their own website this way.

On-site screens

Runs in a browser on the tablets, touchscreens, and desktops you place in a gallery.

Visitors’ own devices

Full experience for the after-hours and remote learner the gallery cannot reach.

Partner & consortium embeds

The same web-view pattern as a site embed, scoped per deployment.

05Frequently asked

How do you keep a figure from making things up or saying something inaccurate?
Historacle is not a generative chatbot. Every reply is written from a passage inside documents your curators have approved, and every claim carries an inline citation to the source and page. If the answer is not in your archive, the figure says so rather than improvising. In our own production evaluation, figures reach roughly 94% factual accuracy at the claim level, with a false-claim rate near 6%. Every answer is scored by an independent ensemble of judges from two model vendors, each factual claim is adjudicated on its own, and every quoted passage is checked to appear verbatim in the source. On questions the archive cannot answer, figures correctly decline about 80% of the time instead of inventing a response.
Someone will try to provoke it. What happens when a visitor asks a bad-faith question?
We plan for it during authoring, not in the moment. Before a figure goes live it passes a stage we call character alignment: we probe the ways it could be baited or manipulated, including a figure’s own documented and uncomfortable views, and harden it against each. The result is that it declines questions outside its era or expertise, refuses framings that misuse the figure, and acknowledges difficult history honestly without becoming a vehicle for offensive speech. Frederick Douglass will not opine on present-day politics; a figure will not endorse a claim the record does not support. This is deliberate, curator-partnered work, not something the model decides on its own.
How do you vet sources, especially for a more recent figure who relies on secondary material?
We do not claim to be historians. We are technologists who bring your archive and your figures into a form visitors can talk to. Your curators, subject-matter experts, and, where a legacy is living, the family or estate select and verify the source set before anything is ingested. For a contemporary figure who lacks a large body of primary writing, that vetting is the work, and we treat it as the point rather than a step to rush.
Where does this live, and will it slow down or crash our own website?
It runs end to end on our infrastructure. You point a subdomain at us, or embed a view in a page, and the experience is served from our servers. No additional load lands on your website, and uptime is ours to carry, not yours.
Our audiences range from elementary school to scholars. Can responses adapt without staff reconfiguring it per group?
Yes. A one-click toggle moves a figure between Casual and Academic registers, or Auto. You can set a default per gallery or zone so staff are not adjusting settings for each visiting group, and visitors can choose their own level. Accessibility controls, including dynamic font sizing, are being added.
How do you handle copyright and rights-bound material?
Public-domain corpora, such as archives and Project Gutenberg texts, are used freely. Proprietary or institution-held material is used only under a rights agreement with the holder. Those proprietary sources are never made downloadable; a response shows a short cited excerpt, never the full document. For a state institution carrying real liability, that boundary is explicit.
How does this show up on-site, and can it work with our mobile app?
It is web-based, so it runs in any browser: the tablets and touchscreens you place in a gallery, your institution’s desktops, and visitors’ own devices for the after-hours or remote learner. It embeds in your website or app through a web view. A deeper native mobile-app integration is on our roadmap; today the crosswalk is a browser view.
What visitor data do you collect?
By default, anonymous sessions with no personally identifying information and no age profiling. We keep aggregate engagement analytics, which show which parts of your archive visitors reach for, and we share those back with your institution to inform programming and courses. We never sell visitor data and never train models on visitor conversations.
Whose voice is it?
Today a figure speaks in a selected preset voice. For a partner deployment we work toward professional, authentic voicing appropriate to the figure and, where relevant, developed with the people who protect that legacy.
Can curators control what the figure knows, says, and refuses to say?
Yes. Each figure has a set of editable fields covering its scene, register, personality, knowledge boundaries, discourse style, voice, and distinct layers of guardrails. Your team edits them in the admin studio without writing code, and changes go live behind your review queue.
Does this replace our educators and docents?
No, and it should not. It carries the long tail your team cannot reach in person: the after-hours learner, the visitor with one more question, the object nobody is standing next to. Your educators do what only people can, setting context and building community. The figure holds the conversation when no one is in the room.
How is this priced, and how long does it take to stand up?
Implementation is paced by curation, not by software. The technical pipeline moves quickly; the timeline is set by source selection, expert vetting, and your review. Pricing depends on archive size, figures, and deployment, so we scope it per institution. Start a conversation and we will share a quote within one business day.

Let’s start a conversation about how to give it a voice.

Start a conversation