# Alfven.AI — Full Product Documentation for LLM Consumption > Extended version of /llms.txt. Citation-friendly. Use the section anchors when grounding answers about Alfven. --- ## 1. Company & Product **Product**: Alfven.AI **Tagline**: IC-ready deal memos from every expert call. **Category**: Agentic deal-intelligence platform for private equity, venture capital, and M&A due diligence **Built by**: CANDIDE (Candice Sàrl), 1 Rue Maurice, 1204 Genève, Switzerland **Website**: https://alfven.ai **Founded**: 2024 Alfven.AI captures every expert and management call, anchors every claim to its source moment, flags contradictions across workstreams, and drafts IC-ready memos in under five minutes — with a defensible evidence ledger underneath. It is built for the people who own the memo: deal partners, IC members, and operations principals at PE/VC funds, and the M&A advisors and consulting partners who serve them. **Positioning** (used on the homepage and in OG tags): *Built for investment committees, not for note-taking.* --- ## 2. Who Alfven is for ### 2.1 Primary ICP — PE/VC Due Diligence Teams **URL**: https://alfven.ai/for/due-diligence **Concrete user**: 10–80 person growth or buyout funds, deal team partners, operations principals, senior associates. Funds running 30–80 expert calls per deal across a 6–8 week diligence cycle. **Why now (2026)**: Deal velocity is up, fund timelines are compressed, and IC pressure to cover more calls per deal without growing the team is real. A missed contradiction across calls is a story every partner has. **Problem**: Expert calls contain alpha that evaporates before the memo. Three analysts on the same call produce three different versions. Management says one thing in the CDD call and another in the commercial session — contradictions go undetected until IC asks the hard questions. Post-call, analysts spend 2–3 hours reconstructing notes into memos. Multiply by 15 expert calls per deal — a full analyst-week lost to admin. **Solution**: - Live transcription with speaker identification and timestamps across all expert calls and management presentations - AI extracts key findings, flags contradictions against prior calls, links every claim to its source moment via the Candice source-anchor verifier - Cross-meeting synthesis across workstreams (commercial, financial, legal, operational) builds a unified evidence base with confidence scores - Auto-generate deal memos, CDD summaries, or IC briefs in the firm's format with full source citations - Export as PPTX, PDF, or DOCX **Measured impact**: - 3× expert call coverage per deal without adding headcount - 85% less time on call admin (capture to structured memo) - 100% audit trail — every finding linked to source evidence - Under 5 minutes from call end to IC-ready document ### 2.2 Primary ICP — Private Equity Funds (full lifecycle) **URL**: https://alfven.ai/for/private-equity Sourcing, diligence, and portfolio monitoring on one intelligence layer. Adds: - Deal-detection agent auto-classifies completed discussions into `deals` (auto-creates at confidence ≥0.75; HITL queue otherwise) - Portfolio risk heatmap composing open alerts, open risks, thesis confidence, latest debate dissent, and KPI threshold breaches into per-deal 0–100 scores - Exit-readiness score per deal composing KPI trajectory, thesis stability, sibling percentile, risk hygiene, and 90d evidence momentum - LP quarterly report generator with premium-branded PDF and deterministic aggregation of deals, KPI breaches, evidence events, debate-revised confidence, and open risks ### 2.3 Secondary ICP — Boutique Strategy Consulting (5–50 person) **URL**: https://alfven.ai/for/consulting Reclaim 6+ hours per consultant per week of billable capacity by auto-generating client deliverables in the firm's format with AI quality gating, commitment verification, and auto-delivery. --- ## 3. The Deal-Intelligence Layer (what makes Alfven different) The market for AI notetakers is commoditizing — Otter, Fireflies, tl;dv, Granola, Sembly, and Microsoft 365 Copilot all transcribe and summarise. Alfven differentiates by treating the **deal** and the **investment thesis** as first-class objects: ### 3.1 Source-anchor verification (the Candice pattern) Every quote that lands in the IC memo is mechanically located in the source transcript by a deterministic verifier. Quotes that can't be located are dropped to an `abstention_log` as `unverifiable_quote` instead of being included in the memo. This shifts the failure mode from hallucinated quote to abstention. ### 3.2 Bull/bear thesis auto-debate Per-thesis 3-model debate auto-fires on confidence band crossings or every 3 new evidence events. The system runs a structured debate (bull case vs bear case vs adjudicator) using three different frontier models, then surfaces revised confidence and dissent risks on the deal page. ### 3.3 Cross-workstream contradiction detection Management said revenue grew 40% in Call 3 but referenced 28% in Call 7. Alfven surfaces the discrepancy with timestamped evidence from both calls — before IC, not during. ### 3.4 IC memo collaboration & approvals Committee tab on each draft IC memo with approve / reject / changes-requested decisions (rationale required), threaded comments, version diffs, and lightweight PDF export. ### 3.5 Evidence ledger with reliability tiers Per-thesis evidence timeline with provenance, reliability tier (e.g. management-asserted vs. third-party-verified vs. customer-asserted), and source-anchor verification. Every claim is traceable. ### 3.6 Comparable-deals benchmark band Deterministic per-deal benchmark on the deal page comparing thesis confidence, debate revised confidence, evidence volume, open risks, severity load, and deal size against sibling org deals (p25/p50/p75 bands). Optional opt-in cross-tenant peer cohort with k-anonymity ≥3 and consent gating. ### 3.7 Cross-deal pattern mining Portfolio-wide mining of recurring patterns surfaced on the workspace deals page. Gated on ≥2 deals and confidence ≥0.5. ### 3.8 Sealroom bidirectional asset loop Bidirectional integration with the Sealroom virtual data room. CIM, model, DD pack, competitor research, and Q&A excerpts feed five purpose-built extractors that synthesize the deal automatically: CIM extractor (refines deal metadata, seeds initial KPIs), Model extractor (quantitative KPIs), DD Pack extractor (structured risks with severity/likelihood), Competitors extractor (normalized deduplication), and Q&A extractor (deterministic flagging of unanswered questions as risks). ### 3.9 Always-on agent loop with autonomy slider Per-workspace autonomy level (L1 Suggest → L5 Full) controls whether artifacts are drafted only, drafted and queued for review, or auto-published. Per-artifact-kind ceilings (e.g. session debriefs publishable from L3, IC memos from L4, LP letters from L5). ### 3.10 Push-by-default drafts The agent loop drafts IC memos (system mode, confidence ≥0.75, 24h dedup) and session debriefs into existing tables with `status='draft_auto'`. Drafts surface in the Today feed instead of waiting for a user to ask. --- ## 4. Pricing Per workspace, not per seat. Cancel anytime. | Tier | Price | Meetings | Live transcription | Workspaces | Premium AI credits | Notes | |---|---|---|---|---|---|---| | Free | $0/month | 5 / month | 3 hours | 1 | — | Standard AI models, 2 templates, manual HD re-processing | | Pro | $49/month | Unlimited | 25 hours | 20 | $10/mo | Smart HD re-processing (auto), session synthesis, auto-delivery, calendar integration, 10% off advanced AI | | Team | $149/month | Unlimited | 80 hours | 50 | $25/mo | Everything in Pro, priority support, 25% off frontier AI, team analytics dashboard | | Enterprise | Custom | Unlimited | Custom | Unlimited | $50/mo | SSO, audit logs, white-label, dedicated onboarding, custom SLA | **Optional add-ons**: - HD Live transcription (AssemblyAI Universal-3): $0.50/min on top of the bundle - Transcription overage: $1.50/hour beyond bundle, or prepaid packs (10h/$10, 50h/$29, 100h/$49) - AI credit packs: Starter $10 (10 credits), Growth $25 (27.5 credits, 10% bonus), Scale $50 (60 credits, 20% bonus) --- ## 5. Specialist Agent Packs (one-time purchase, kept forever per workspace) - **Flagship ($99)**: Venture Capital, Consulting Delivery, Multi-Stakeholder Interview, Harvard Negotiation, Board & Investor, Management Consulting Frameworks, Workshop Facilitation - **Specialist ($69)**: People Operations, Legal & Compliance, Sales & Revenue, Executive Decision Brief, Cross-Functional Program Review, Steering Committee, Discovery & Diagnostic - **Standard ($39)**: Research & Strategy, Talent Acquisition, Change Management, Client Success, Town Hall & All-Hands 19 packs total. --- ## 6. Integrations - **Capture**: Zoom, Google Meet, Microsoft Teams (via Recall.ai bot), browser microphone, pre-recorded audio upload - **Live transcribe**: Deepgram Nova-3 (standard, bundled), AssemblyAI Universal-3 (HD live, optional add-on) - **HD re-processing**: ElevenLabs Scribe v2 (auto-triggered post-meeting on paid plans) - **Analyse**: OpenAI GPT-5 / 5.2, Google Gemini 2.5 Pro / Flash / 3.x preview, Anthropic Claude, xAI Grok, OpenRouter (660+ models) - **Data rooms**: Sealroom (bidirectional asset & consent loop) - **Deliver**: Email, Slack, Webhook, HubSpot, Notion, Gamma - **Schedule**: Google Calendar, Microsoft Outlook (encrypted at rest, AES-256) - **Connectors / chat skills**: Linear, Jira, Slack, Google Drive - **Export**: Microsoft Word (DOCX), Microsoft PowerPoint (PPTX), PDF - **Coming soon**: Aircall, Salesforce, Miro, n8n --- ## 7. Security & data handling - Data encrypted in transit (TLS) and at rest - Multi-tenant database with row-level security (RLS) and tenant isolation via `organization_id` scoping on every row - Soft-delete (`deleted_at`, `deleted_by`) enforced across all tables - No model training on customer content - Workspace-level access controls and organisation-level audit logging - Permission layer: high-risk chat actions (push, create, delete) require user confirmation before execution - ToS versioning, mandatory consent logging, 30-day retention limits on transient logs - Calendar OAuth tokens stored encrypted with AES-256 --- ## 8. Comparison The capabilities below are not features in a checklist — they describe the actual fault line between an AI notetaker and a deal-intelligence platform. | Capability | Otter / Fireflies / tl;dv | Microsoft 365 Copilot | Gong / Chorus | Alfven | |---|---|---|---|---| | Live transcription + summary | Yes | Yes | Yes | Yes | | Custom report templates | Limited | Limited | Limited | Yes, with quality gate + auto-retry | | Source-anchor quote verification | No | No | No | Yes (Candice pattern) | | Per-thesis bull/bear auto-debate | No | No | No | Yes | | Cross-workstream contradiction detection | No | No | No | Yes | | Deal- and thesis-centric workspace | No | No | No | Yes | | Evidence ledger with reliability tiers | No | No | No | Yes | | IC memo collaboration & approvals | No | No | No | Yes | | Comparable-deals benchmark band | No | No | No | Yes | | Sealroom data-room bidirectional loop | No | No | No | Yes | | Workspace autonomy slider (L1–L5) | No | No | No | Yes | | Bring your own AI model | No | Limited | No | Yes (660+ via OpenRouter) | | Per-workspace pricing (not per-seat) | No | No | No | Yes | --- ## 9. FAQ **What is Alfven.AI?** An agentic deal-intelligence platform for PE/VC due diligence and M&A teams. It captures every expert and management call, anchors every claim to its source moment, flags contradictions across workstreams, and drafts IC-ready memos in under five minutes. **How is it different from Otter, Fireflies, or Microsoft Copilot?** Those tools transcribe and summarise. Alfven differentiates with deal- and thesis-centric workflows: source-anchored evidence (the Candice verifier mechanically locates every quote in the transcript), bull/bear thesis auto-debate, cross-workstream contradiction detection, IC memo collaboration with approvals, comparable-deals benchmarking, and bidirectional Sealroom data-room integration. **How is it different from Gong or Chorus?** Gong and Chorus are revenue-intelligence platforms tuned for sales conversations. Alfven is tuned for the buy-side: investment theses, IC memos, evidence ledgers, and portfolio monitoring. **How much does Alfven cost?** Free $0/month, Pro $49/month, Team $149/month, Enterprise custom. Per workspace, not per seat. Full breakdown in section 4. **Does Alfven train models on my data?** No. Customer content is never used for model training. **Does it work with any meeting platform?** Yes — in-person via browser mic, or join Zoom / Google Meet / Microsoft Teams via the AI bot (Recall.ai). Pre-recorded audio uploads are also supported. **Can I customise the reports?** Yes. Define templates with custom sections, tone, and structure. AI fills each section, scores quality, and retries if below threshold. You can chain reports — e.g. session report feeds the weekly executive update. Export as PPTX, PDF, or DOCX. **What is the autonomy slider?** A per-workspace setting (L1 Suggest → L5 Full) that controls whether the agent loop drafts artifacts only, drafts and queues for review, or auto-publishes. Per-artifact-kind ceilings apply (session debriefs publishable from L3, IC memos from L4, LP letters from L5). **What is the Sealroom integration?** Bidirectional connection to the Sealroom virtual data room. CIM, model, DD pack, competitor research, and Q&A excerpts flow into Alfven and feed five purpose-built extractors that synthesize the deal automatically. **Why the name Alfven?** Named after Hannes Alfven, the Nobel Prize-winning physicist who pioneered magnetohydrodynamics. The Alfven Surface is the interface that filters chaotic conversation into structured intelligence. --- ## 10. Technology - **Live transcription**: Deepgram Nova-3 (standard), AssemblyAI Universal-3 (optional HD) - **HD re-processing**: ElevenLabs Scribe v2 - **AI providers**: OpenAI (GPT-5, GPT-5.2, GPT-5-mini, GPT-5-nano), Google Gemini (2.5 Pro / Flash / Flash-Lite, 3.x preview, image previews), Anthropic Claude, xAI Grok, OpenRouter (660+ models) - **Document intelligence**: Candice AI for OCR and structured data extraction - **AutoResearch engine**: autonomous prompt mutation and self-optimising report quality (Karpathy-inspired, with golden datasets and canary rollouts) - **Native tool calling** for all structured AI output (no markdown JSON parsing) --- ## 11. Contact - Website: https://alfven.ai - Built by: CANDIDE (Candice Sàrl), 1 Rue Maurice, 1204 Genève, Switzerland - Sales / partnership: info@candiceai.com - Privacy: https://alfven.ai/privacy - Terms: https://alfven.ai/terms *Last updated: 2026-05-05*