Knowledge assistant

    Answers your teamcan actually cite.

    Ujiboo Agent turns your documents — uploads, intranet pages, public portals — into a conversational knowledge base. Every answer arrives with its sources attached. No hallucinations, no guesswork.

    GDPR · EU residencyEvery answer citedBring your own LLMSelf-hosted available
    The conversation

    One workspace. Every question answered.

    A chat surface that remembers context, narrows scope when you ask it to, and never answers without showing its work. Built for the people who'd otherwise spend half their day searching PDFs.

    • Real conversation, not search

      Follow-ups stay in context. Ujiboo Agent keeps the thread even across hours and topics.

    • Cited from your own docs

      Every claim links to the source it came from. No invented references.

    • Knows what's in scope

      Per-user scopes (public · shared · office · private) shape every retrieval.

    Anatomy of an answer

    Every answer brings receipts.

    An answer without a source is a guess. Ujiboo Agent retrieves the exact passages it used, scores them by relevance, and shows them right next to the response. If you don't trust it, click and read.

    • Verbatim citations

      Each cited source surfaces the exact passage the Agent used, with page reference and relevance score.

    • Grounded by design

      If the answer isn't supported by retrieved chunks, the Agent says so. No silent guessing.

    • Versioned, auditable

      Every answer keeps the prompt version, model, and source set. Easy to replay weeks later.

    • Feedback that trains itself

      Thumbs-up / down flows back into evaluation runs. Quality keeps improving with use.

    See the knowledge base
    Knowledge base

    Your documents, queryable in minutes.

    Drop in a folder of PDFs or wire up a public portal. Ujiboo Agent parses, chunks, and indexes everything in the background. Visibility scopes are enforced at retrieval time — not bolted on later.

    • Drop in anything

      PDFs, Word docs, web pages, intranet exports. Text is parsed, chunked, and indexed automatically.

    • Auto-refreshed sources

      Scrape public portals on a schedule. Only the chunks that changed get re-embedded.

    • Scope-aware retrieval

      Public, company-shared, office, private. Every retrieval respects the user's scope at runtime.

    Analytics

    See what your team actually asks.

    Every conversation is logged with latency, token usage, cost, and feedback. Top intents surface gaps in the knowledge base. One chart, full ROI story.

    • Volume, latency, cost

      Every conversation logged with response time, tokens used, and per-turn cost. Defensible ROI.

    • Helpfulness, tracked

      Thumbs up / down per answer rolls into a single quality score over the last 7 / 30 / 90 days.

    • Drill into a single turn

      Click any conversation to see the retrieval, the prompt, the model used, and the bill it produced.

    Channels

    Same brain, three places to live.

    Internal app for your team. Embedded widget on your public site. A REST API for everything else. One knowledge base, one prompt, one set of guardrails — three surfaces.

    • Internal dashboard

      Full-featured chat for staff. SSO, scopes, prompt studio, analytics.

    • Embed widget

      One script tag, any site. CAPTCHA, theme, scopes inherited from the workspace.

    • Plain REST API

      Stream chat completions with citations. JSON in, JSON out — wire it into anything.

    See the prompt studio
    Prompt studio

    The prompt is a real artifact.

    Tone, formatting, refusal rules, source guardrails — all edited in one place, versioned, evaluated, then published. You change the assistant's behavior without redeploying anything.

    • Prompt as code

      Per-phase system prompts (retrieval, generation, title) edited in one place.

    • Versioned & revertable

      Every change is a tracked version. Roll back in one click if a release misbehaves.

    • Evaluated before publish

      Run a draft against the last 100 questions. See helpfulness, citation rate, hallucination delta.

    • Bring your own LLM

      Azure OpenAI, OpenAI, Bedrock-ready. Choose the model per phase, see the cost.

    Access & security

    Your knowledge. Your perimeter.

    Ujiboo Agent is built so your data stays where it's supposed to. Choose your model, choose your region, choose whether it's hosted or self-hosted. Auditable end-to-end.

    • EU residency

      Data stays in EU regions you approve. No cross-region replication, ever.

    • Self-hostable

      Run the Agent on your own infrastructure — full control of model traffic and storage.

    • Bring your own model

      Azure OpenAI, OpenAI, Bedrock. Use your tenant, your credentials, your billing.

    • SSO + roles

      OIDC / SAML / SCIM. Granular roles map to scopes and admin capabilities.

    • Per-scope visibility

      Documents are tagged at ingest. Retrieval enforces scope server-side, every turn.

    • Full audit trail

      Every prompt, every retrieval, every answer — logged. Replay any conversation.

    GDPR-aligned·DPA on request·Audit log · 12-month retention·Encryption at rest + in transit·Daily backups
    Discovery starts here

    Let's give your team answers they can trust.

    Bring a folder of documents and a few real questions. We'll wire up a working Ujiboo Agent in a discovery call and show you the answers — with citations — live.

    • Tailored configuration
    • Onboarding in days, not months
    • Self-hosted available