01 · Ingest
Omnichannel in
Email, web chat, WhatsApp and SMS arrive through one channel gateway and are normalised into a single event.
The platform
From patron message to approved answer.
One pipeline turns an inbound question into a grounded, human-approved reply — across every channel, grounded in your own systems.
The decision engine
Every request flows through one intelligent pipeline.
02 · Classify
AI understands
An intent classifier labels the request — content search, circulation, ILL, fines, DRM or support — with a confidence score.
03 · Act
The agent drafts
Tenant rules pick the agent; it retrieves context (RAG + your connectors) and drafts a grounded reply — optionally on a local LLM.
04 · Review
A human approves
The draft lands in the Cases queue. Staff edit, check citations, and approve — nothing is auto-sent.
05 · Send
Delivered & logged
Approved replies go out on the patron’s own channel; every step is captured in a tamper-evident audit trail.
Grounded in your systems
Not a chatbot. A grounded operator.
Answers are built from your catalogue and your knowledge — with the option to keep inference entirely local.
Your connectors
Live reads from your ILS/LSP and 25+ research sources feed every draft with real holdings and citations.
Hybrid knowledge (RAG)
A knowledge layer retrieves relevant context so replies are specific to your collection and your policies.
Local or hosted models
A model gateway routes to a local LLM (LM Studio) in privacy-first mode, or to a provider you choose.
Omnichannel
One inbox for every channel.
Meet patrons where they already are — every channel runs the same review-before-send flow.
| Channel | Availability | How it works |
|---|---|---|
| Production-wired | Inbound + outbound over Gmail and Microsoft 365 (Graph) with OAuth — no passwords stored. | |
| Web chat | Embeddable | A chat widget that drops onto your site and feeds the same pipeline. |
| Via Twilio | Two-way WhatsApp through Twilio when configured. | |
| SMS | Via Twilio | Text-message support on the same review-before-send flow. |
The surfaces
Built for the people who run it.
Everything staff and patrons touch, on one platform.
Surfaces
- Admin Console
- The staff control plane: the Cases queue, Connector marketplace, Agent Studio, the Rules decision engine, Prompt Studio and a Runtime Explorer for traces and audits.
- Patron Portal
- A patron-facing surface to ask questions and submit support or research requests straight into the pipeline.
- Analytics Dashboard
- Usage rollups and operational analytics drawn from the shared audit log.
- Chat Widget
- An embeddable web-chat entry point for your own site.
Questions
The short version.
Can it run fully on-premise?
Yes. The platform can run inside your own network and route inference to a local LLM via the model gateway, so patron data and conversations never have to leave the building.
Which library systems does it connect to?
Ten ILS/LSP vendors (Koha, FOLIO, Evergreen, Alma, Sierra, Polaris, Symphony, WorldShare, Destiny, SLiMS), five standard protocols (Z39.50, SRU, SIP2, NCIP, OAI-PMH) that reach many more, plus DSpace and 25+ open research sources.
Is a patron ever answered without a person checking?
No — by design. Every AI draft lands in the Cases queue for a staff member to edit and approve. Governance rules can require citations and block auto-send.
What about payments and e-book licensing?
Fines are paid through Stripe hosted checkout (PCI-safe), and e-book access is issued as Readium LCP licenses.