For Archives, Libraries & Heritage Collections

Open the boxes nobody opens.

Historacle brings the people inside your boxes into the reading room. Researchers sit with the correspondent, hear the oral-history subject in their own voice, walk through the migration the collection documents. Provenance preserved, donor restrictions honored, every answer cited to box, folder, and page.

Start a conversationSee how it works

01The opportunity

Much of the material in U.S. archives is described only at the collection level, if at all, and roughly half has no online presence (OCLC, 2010). Researchers who would use the material can’t find it; the public doesn’t know it exists. Meanwhile the question coming most often to your reference desk is no longer “how do I find a record?”. It is “what does this collection say about my grandmother, my town, my movement, my people?”

That question is one your collection can answer. But it requires a layer of access between the catalog and the reader that doesn’t exist yet. We built that layer.

02How it works

A deployment is not a chatbot on a finding aid. Researchers sit with the figure, in a scene drawn from your holdings, and the figure speaks in their own voice. Where your collection supports it, the oral-history subject is heard in the original recording, the photograph is shown in scene, the place in the record is walked through.

01

Researchers sit with the people in your collection. A founder. A correspondent. The subject of an oral history, speaking in their own voice, restored from the recording you hold. The figure answers only from documents you have approved, with full citations to box and folder.

02

Walk a researcher through the strike, the migration, the town’s evolution, the community’s story, guided by a figure within your holdings, narrating the world the collection documents.

03

Sit in on a multi-voice conversation across collections, figures from different boxes, eras, and donors, so the user encounters the same event from the angles your archive actually holds.

04

Convert any conversation into research aids, finding-aid drafts, transcription support, and outreach materials. Where your collection supports it: hear the oral history in the subject’s own voice, see the photograph in scene, walk through the place the record describes.

03The hard part, done right

Archival authority rests on a single principle: every claim has to lead back to a source. That principle has been under quiet assault from AI tools that summarize without citing, cite without grounding, or simply make material up.

Historacle is built around the archival principle. Every response is grounded in documents your team has approved, and every claim ships with a citation: collection, box, folder, page. When a question reaches beyond what your holdings can answer, the figure says so. Donor restrictions, access conditions, community protocols: enforced at the document level. The figure cannot speak from material you haven’t released to it.

  • Grounded in your collections only. No external corpus. No general-web bleed.
  • Every answer cites the collection, box, folder, and page. Provenance preserved end-to-end.
  • Donor restrictions honored by controlling the source set. The figure cannot speak from what isn’t released to it.
  • Community-owned and culturally restricted material stays out unless you include it. Your protocols, never improvised by ours.
  • Your collections stay yours. No cross-tenant training, no third-party fine-tuning, 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 catalog or finding-aid site. A few of the ways institutions put their figures in front of researchers today:

  • Your own catalog or finding-aid site: your URL, your branding, embedded the way a partner institution runs its figure on its own website
  • Direct web on historacle.ai: for collections whose figures are publicly accessible
  • Mobile: works in any browser, opened from the reading room or from home
  • Reading-room and on-site use: accessed from existing desktops, laptops, or tablets the institution already runs
  • Partner and consortium embeds: same pattern as the catalog embed, scoped per deployment
  • After-hours and remote research: full-feature web for everyone the reading room can’t reach

Built for archives

Historacle is built for archives, libraries, and historical societies. Our early pilots run across museums, historic sites, and music-history collections, and the same pipeline carries directly to archival holdings. To bring your collection into the reading room, start a conversation.

05Frequently asked

How do you prevent the AI from misrepresenting an archival source or making things up?
Every response is grounded in documents your team has approved, using a search system that finds the right passage by meaning and by exact terms, then writes the answer from that passage alone. Every claim carries a citation to the collection, box, folder, and page. If the answer is not in your collections, the figure says so. Independent 2025 studies found that even AI systems pulling from source documents still fabricate on a meaningful share of factual questions; we close that gap with strict source-locking, scope guardrails, and an explicit “I don’t have that in the record” refusal pattern.
Whose data trains the AI? Will our collections be used to train models for other clients?
No. Your holdings live in an isolated index that only your characters can read. We do not fine-tune base models on your data. We do not share, sell, or expose your collections across organizations. Other Historacle clients cannot retrieve or be influenced by your material. If you leave, your data is exported and our copies are deleted when the engagement ends.
How does Historacle integrate with our finding aids, catalog, and DAMS?
We ingest a wide range of material: documents, books, newspapers, images, and audio, in the formats your team already uses, including EAD finding aids, MARC catalog records, PDF, Word, plain text, Markdown, and structured CSV or JSON. Our document pipeline runs an eight-phase process (conversion, chunking, metadata, sanitization, validation, embedding, indexing, verification) and supports incremental updates as your collections grow. We work with your catalog and DAMS exports; deeper API integrations are scoped per pilot.
Who is responsible for the accuracy of what the AI says about archival material?
Accuracy is a partnership: your archival authority sets the boundaries, and the platform is built to hold the figure inside them. Your archivists define the source set, the persona and scope, the topics the figure will engage, and the refusals it must observe. The figure draws only from material you have released to it, cites it on every claim down to box, folder, and page, and declines when the source does not support an answer. Administrative actions are captured in the audit log, so archival authority stays with you.
How do you handle donor restrictions, access conditions, and community-owned material?
You control it by controlling the source set. The figure can only draw from the documents you release into its index, so restricted or community-owned material simply is not ingested unless you choose to include it. You can also scope a figure to refuse whole topics by guardrail, for example, "this figure does not discuss material restricted to ceremony." For material that needs conditional handling, we work through your access protocols together during setup.
What researcher or visitor data is collected, and how is it stored?
By default: anonymous session traces with no personally identifying information, aggregate engagement analytics, and only the conversation transcripts researchers explicitly save. Researchers can request deletion at any time, and we delete data when the engagement ends. We never sell researcher data and never train models on researcher conversations.
Can archivists control what the figure knows, says, and refuses to say?
Yes. Each character has twelve editable fields covering scene description, historical style, modern style, personality modes, knowledge boundaries, casual and academic discourse instructions, voice instructions, and three distinct layers of guardrails. Archivists edit them in the admin studio without writing code; changes go live behind your review queue.
What does implementation actually look like, and how long?
Implementation is paced by curation, not by software. The technical pipeline moves quickly; the timeline is set by source selection, archivist review, and the vetting your holdings deserve. You provide the source material and archival direction; we handle the pipeline, hosting, and the figure’s voice.
Does this serve casual researchers, scholars, or both?
Both. Curators set the discourse level per character: a casual public-facing figure that answers in plain language, an academic figure that cites in full archival form, and everything between. The same archive can power a kiosk for a tenth grader and a research aid for a dissertator: different characters, different voices, the same sources underneath.
How is this priced?
We don't publish public pricing because deployment scale, collection size, and researcher volume vary widely across institutions. Pilots are scoped for cost predictability so your spend doesn't fluctuate with conversation volume. Start a conversation and we'll share a quote within one business day.

Let’s open a box.

Start a conversation