Truvyo is an evidence-grounded AI support engine. Two agents — Customer Agent and Technical Companion — answer from a verified, SKU-specific knowledge base and cite every claim. No hallucination, no extract endpoint, no sales motion.
A small, self-contained AI support platform built on Cloudflare. The product surface fits on a single architecture diagram.
| Component | What it does |
|---|---|
| Customer Agent | End-user-facing. Claude Haiku 4.5. Plain-language symptom triage. Always cites sources. |
| Technical Companion | Installer/operator-facing. Claude Sonnet 4.6. Multi-symptom diagnostics, install sequencing, escalation paths. |
| Hybrid retrieval | Vectorize (semantic) + D1 FTS5 (BM25) + Reciprocal Rank Fusion. Top-5 articles per turn. |
| Shared Knowledge Library | Curated manufacturer and vendor documentation. Compounds across every tenant. |
| Tenant overlay | Per-tenant private KB. Internal SOPs, brand voice, policy. Never extractable. |
| Cross-check audit | Every published article passed an automated audit against authoritative sources before serving. |
| Disposition billing | Every activated chat creates a disposition row; only the ones the agent resolved are billable. |
| Auto-update pipeline | Quarterly ingest of new vendor docs + community evidence + audit gate. Daily aggregation of feedback and gaps. |
Five rules the engine is built around. Everything else falls out of them.
If a claim cannot be grounded in an audited source, the agent does not make the claim. There is no "confident guess" mode and no fallback to model parametric memory. When retrieval comes up empty, the agent says so and escalates.
Choice of LLM is a config line. The hard work is the governance — retrieval filters that only surface published articles, an audit pipeline that benches unverified claims, disposition logic that ensures billing only fires when the work was done. Engineering time goes there.
Documentation is the salesperson. Pricing is public. Architecture is public. Sign-up is one form. If the trial doesn't convert you, no demo would have.
Tenant content goes in. It does not come out as a file. No bulk export endpoint exists; no admin route returns article bodies; no UI exposes full content. Verified in code, not just terms of service.
If Truvyo couldn't ground the answer and the conversation escalated to a human, Truvyo didn't do the work — and doesn't charge for it. Industry-standard contact-center terminology: dispositions. Billable dispositions are the ones the agent resolved.
| Headcount | Small, engineering-heavy team. |
| Sales motion | None. No outbound, no SDRs, no "schedule a demo" between buyer and product. |
| Pricing model | Public, transparent, three components: basic platform fee + pooled dispositions + $45/seat. Full pricing. |
| Trial | 7 days, no credit card, self-serve via the website. |
| Deployment | Cloudflare Workers + D1 + Vectorize. Multi-region; data resides in the tenant's selected region. |
| LLM provider | Anthropic (Claude Haiku 4.5 and Sonnet 4.6). Provider isolated to a single module; swap is a one-line change. |
| Data policy | No training on tenant data. SHA-256-hashed API keys. Per-tenant row isolation. No cross-tenant queries possible. |
| Support | Email-first. Docs are the first line; support@truvyo.ai when docs don't cover it. |
In order of priority. The order is intentional — we ship in the order things create the most value.
| For | |
|---|---|
| General | hello@truvyo.ai |
| Engineering questions | eng@truvyo.ai |
| Support & billing | support@truvyo.ai |
| Security disclosures | security@truvyo.ai |
| Press | press@truvyo.ai |
| Careers | careers@truvyo.ai |
7 days free. No card. No sales call. 30 seconds to a trial API key.
Start free