# Your company's brain. Indexed. Cited. Always current.

## 01 · Every question, answered at the moment of work.

Your employees ask questions. Your agents query in loops. Both need answers in milliseconds — from every document, video, thread, and system in your stack. Quelvio retrieves all of it. PDFs, videos, audio recordings, Slack threads, wiki pages, meeting recordings, presentations, spreadsheets — 14+ formats, semantically understood. Authority-ranked, not keyword-matched.

> The answer that used to take 6 emails, three Slack DMs and a meeting — now takes one query.

Connect your sources — Drive, SharePoint, Confluence, Notion, Slack, Teams, S3 — and search across your entire knowledge base from day one. No training period, no manual tagging, no six-month implementation.

## 02 · What your organization actually knows.

Search works from day one. As your content grows, Quelvio builds something deeper — a continuously updated model of where authority sits, which knowledge is decaying, and where you're one departure away from losing critical context. Intelligence the system can act on. Diagnostics you can show your board.

• -Know who actually knows what: Identify the highest-authority sources for any domain — ranked by actual contribution, not job title. The signal across documents, code, decisions, and threads.
• -Catch decay before it becomes a problem: A diagnostic across coverage depth, content freshness, and consistency. Surfaces where knowledge is decaying before it becomes a problem.
• -See what you're one departure from losing: Domains where critical knowledge concentrates in one source. Identifies content with no equivalent backup — and the closest alternates already in your system.
• -Find the questions you can't answer yet: Per-domain scoring against actual query patterns. Identifies specific topics where employees and agents are searching but coming up empty.

Included with every Enterprise contract. Intelligence is what makes search indispensable — gating it as a premium add-on defeats the purpose.

## 03 · Built for humans and agents.

Quelvio is the only context layer where employees and AI agents draw from the same brain. Employees get cited answers inside Slack, Teams, Claude, Cursor, ChatGPT, and the dashboard. Agents get the same ranked, permission-aware context via MCP and REST — ready for every retrieval inside their loops.

• -For humans: Cited answers inside Slack, Teams, Claude, Cursor, ChatGPT, and the dashboard. Every claim is permission-aware and traceable to the original author and source.
• -For agents: Authority-scored, cross-referenced, freshness-aware context delivered via MCP and REST. Ready for every retrieval inside an agent loop — copilots, autonomous agents, or your own tooling.
  • -Connect via MCP: https://mcp.quelvio.com/http
  • -Install the CLI for terminal agents: npm i -g @quelvio/cli

  Pre-built skill files for Claude Code, Cursor, OpenAI Codex CLI, and generic MCP runtimes are available in our public [github.com/Quelvio/quelvio-agent-skills](https://github.com/Quelvio/quelvio-agent-skills) repo — install once per agent and Quelvio invocation becomes native.

  • -[AI agents integration guide](/docs/ai-agents)
  • -[MCP server reference](/docs/mcp)
  • -[CLI reference](/docs/cli)

## 04 · Knowledge that reasons over time.

Your organization's knowledge doesn't stay still. Documents are updated. Decisions are reversed. Senior engineers leave. A policy from 2024 quietly gets superseded by a Slack thread in 2026. Most retrieval systems treat every source as equally valid. Quelvio doesn't.

• -Cross-references that name the disagreement: When two sources contradict each other, the right answer is not to average them or pick silently. Quelvio names the conflict explicitly — 'On B2C vs B2B strategy, [3] supersedes [1]' — and shows the reader how the organization's thinking evolved. Every source's relationship to every other source is tracked at ingest time: agreement, disagreement, complementary, superseding.
• -Lifecycle awareness, not just freshness: Some content is permanent — canonical policies, foundational principles, code in production. Some is temporary — sprint plans, drafts, status updates. Some must be forgotten — deprecated workflows, right-to-be-forgotten requests, retention-policy expirations. Quelvio classifies content by lifecycle and ranks accordingly. Stale content gets demoted. Current thinking surfaces.
• -Supersession, named explicitly: When a recent high-authority source updates an older one, Quelvio surfaces both — and tells you which is current. Answers cite both the original and the update, with the supersession relationship explicit in the response. No silent overwrites. No averaged truths. You see how the organization's understanding evolved.
• -Refusal as a feature: When the corpus does not contain the answer, Quelvio says so — not invents one. Confidence is calibrated to evidence. The system is willing to refuse when uncertain, which is the difference between a knowledge system and a hallucination machine. Procurement reviewers and security teams notice this immediately.

[Why this matters] Generic RAG retrieves whatever indexed first. Glean ranks by recency and popularity. Quelvio ranks by who actually knows what — and tells you when sources disagree. This is what compounds as a moat over time: the cross-reference graph, the authority signal, the lifecycle awareness. Your AI agents will be only as good as the knowledge they retrieve from. Quelvio is the only context layer built for that.

This is the architecture that compounds. As your content grows, the brain doesn't just have more to retrieve — it understands more about what's authoritative, what's current, and what's been quietly superseded.

## 05 · Your content, everywhere it lives.

• -Google Drive
• -S3 / GCS / Azure Blob
• -Upload API + UI
• -SharePoint / OneDrive
• -Confluence
• -Notion
• -Slack
• -Microsoft Teams
• -Box
• -Dropbox
• -Salesforce
• -Custom webhook

Every connector follows the same pattern: authenticate, select scope, sync. Permissions from the source system are respected at query time — employees only see content they have access to in the original tool.

## 06 · Every format, fully understood.

Text extraction is the baseline. Quelvio goes deeper. Diagrams are embedded as visual vectors and retrievable by content, not just described. Video transcripts carry per-speaker attribution, with cross-modal frame selection that captures the exact slide on screen when the engineer explains the architecture. Tables preserved as structured data. A 1,000-page engineering manual with 200 diagrams is fully understood and searchable by visual content, not just OCR'd.

• -PDF: Text, tables, OCR for scans, vision for diagrams
• -Video / Audio: Whisper transcription, speaker diarization, cross-modal frame selection — captures the frame when the speaker explains it
• -PPTX: Speaker notes as primary content, chart/diagram vision
• -Confluence / Notion: Structured blocks, macros resolved, properties as metadata
• -Hierarchical chunking: Child chunks for precise retrieval, parent chunks for full-context synthesis. Contextualized embeddings preserve document-level meaning at the passage level — answers cite the exact paragraph without losing the surrounding logic.

## A core conviction: Knowledge has a source — always.

Every claim Quelvio surfaces is traceable to the document, the author, the timestamp. No paraphrases without provenance. No synthesis without citation. No answers when the corpus is silent.

## 07 · Per-seat, sized to your contract.

Enterprise is custom-priced per contract with a 100-seat minimum. Pricing scales linearly with your team — including AI agents, which count as seats with their own daily kT pool. Sized to your workload, not a fixed tier.

[See full plan breakdown](/pricing)

## 08 · Where your team already works.

Employees won't switch to a new tool. Quelvio meets them where they are.

• -Dashboard: Search, knowledge map, expert rankings — via SSO at {company}.quelvio.com
• -API: REST API for direct integration — query, source provenance, domain listing, and ingestion endpoints. Embed Quelvio into internal portals, CI/CD bots, custom tools, and AI agent pipelines. OpenAPI spec at https://api.quelvio.com/openapi.json.
• -MCP server: Connect Quelvio to Claude, GPT, Cursor — your AI tools search internal knowledge
• -CLI: quelvio query from your terminal. For Claude Code, OpenAI Codex CLI, custom scripts, and CI/CD pipelines. Authentication via Personal Access Tokens, OAuth login (coming v0.2.0), or service accounts.
• -Slack bot: @quelvio in any channel. Attributed answers in-thread. Zero integration.
• -Teams bot: @quelvio in Teams. Same experience for Microsoft orgs.

## 09 · Security & compliance.

Procurement reviewers print this page. Built for regulated environments, with zero data retention from LLM providers, tenant isolation at three layers, and a Consent Transparency Log that's independently verifiable. Some certifications are in progress — we are transparent about where we are.

• -Encryption: All data encrypted at rest with Quelvio-managed keys (AES-256 via AWS KMS). TLS 1.3 at public endpoints. API keys hashed, never stored in plaintext.
• -Data residency: EU (Frankfurt) and US East (Virginia) on roadmap. Standard DPA available on request. GDPR-aligned.
• -SOC 2: Type I audit engagement Q3 2026. Type II observation begins Q1 2027.
• -Query privacy: Query text is hashed before storage; only audit metadata is retained. LLM providers operate under zero-data-retention contracts.
• -Tenant isolation: Three-layer separation: PostgreSQL row-level security, Qdrant per-tenant collections, S3 per-tenant prefixes. Deletion cascades within 30 days.
• -Permission-aware retrieval: Documents inherit access controls from the source system. Employees only see results from documents they had access to in the original tool. Verified at query time, never cached.
• -No training: Enterprise content is never used to train models. Inference APIs only. Deletion cascades within 30 days, with deletion certificate on request.
• -Consent Transparency Log: Append-only ledger of consent events, Merkle-anchored with daily root hash. Independently verifiable.