diff --git a/docs/00-vision.md b/docs/00-vision.md
new file mode 100644
index 0000000..9622a9b
--- /dev/null
+++ b/docs/00-vision.md
@@ -0,0 +1,71 @@
+# Vision
+
+## What Blocao is
+
+Blocao is a **distributed edge AI platform** for sites that need video and audio forensics with full data sovereignty. It runs on commodity hardware (OpenWrt routers + RK3588-class compute), uses open-source AI engines (Frigate, Whisper, pyannote), and stores everything on-site by default. The hub, when used, runs on operator-owned EU sovereign infrastructure — never on a hyperscaler.
+
+The product is sold as a **vertically integrated platform**: hardware, firmware, console, AI stack, optional fleet management. Customers own their data, their encryption keys, their audit trail.
+
+## What Blocao is not
+
+- **Not a SaaS**. Customers don't pay per-camera-per-month for cloud features. The cost model is hardware + optional support contract.
+- **Not a re-skin of an existing VMS**. The product reaches deep into the network layer (router, VLANs, MQTT broker, GitOps) and the AI layer (Frigate config, embeddings, voiceprints) — it owns the full stack.
+- **Not built on hyperscalers**. AWS, GCP, Azure are not in the critical path. They can be optional sinks for backups but never the source of truth.
+
+## Why this matters now
+
+Three trends converge:
+
+1. **Sovereignty is becoming non-optional**. EU regulators are tightening on extraterritorial data flows post-Schrems II. Argentina's AAIP and similar bodies in LATAM are following. Customers can't put video of identifiable people on US-jurisdiction clouds without legal exposure.
+2. **Edge AI hardware became cheap**. RK3588 with 6 TOPS NPU costs <€200. Two years ago this required a Jetson at €700+. The cost case for "ship raw video to cloud and let GPUs in there process it" no longer holds.
+3. **Open-source AI is competitive**. Frigate, Whisper, pyannote, CLIP, and small LLMs do for video and audio what was vendor-locked SaaS three years ago. The integration work — making them production-grade and operator-friendly — is the value-add.
+
+The window where a sovereign, edge-first, open-source-stack alternative is differentiated is **now**. In three years, every VMS will claim "we run on edge too" but they'll still tether to vendor cloud.
+
+## Who buys this
+
+Three customer profiles, in order of validation:
+
+1. **Mid-sized industrial / agropecuario sites** (LATAM, especially Argentina). They have many cameras, low connectivity, sovereignty concerns mostly driven by IT skepticism of US cloud. They want PoE-equivalent solutions over long distances (HaLow story).
+2. **Small municipalities and event perimeters**. Public-space surveillance with strict retention laws (CABA Ley 5.688, BCRA for banking). They need on-site retention with auditable policies.
+3. **EU SMBs and multi-site retailers**. GDPR exposure is the trigger. Need fleet management across <50 sites, with consolidated investigations and evidence-grade exports.
+
+We're **not** initially targeting:
+
+- Large enterprises with existing Genetec/Milestone deployments — switching cost too high.
+- Police / national security buyers — long sales cycles, evidence chain has to be fully done first.
+- Pure consumer market — the product is too operator-heavy.
+
+## Three-year arc
+
+**Year 1 (2026)**: ship MVP, land 5-10 reference customers in Argentina + Spain. Validate the demo path. Refine based on real ops feedback.
+
+**Year 2 (2027)**: scale to ~50 sites under management. Add transcript forensics as second product line. Open partner channel (resellers self-host hubs).
+
+**Year 3 (2028)**: cross 200 sites. Evidence chain v2 production-ready. Expand to 3-5 EU markets. Optional: hardware partner deal for dedicated Blocao Cell SKU.
+
+After year 3, the strategic question is whether to remain an independent vendor, partner with a security integrator, or be acquired by an open-source-friendly platform vendor (Hetzner, Cloudflare for the sovereignty thesis, or a major OEM for the hardware vertical).
+
+## Bet structure
+
+The bet is that:
+
+1. **Sovereignty + open-source + edge AI** is a defensible positioning (each one alone isn't, the combination is).
+2. **The integration work is the moat**. Anyone can install Frigate and Mosquitto. Making them work end-to-end with GitOps, evidence chain, fleet management, and an operator-grade UX is the hard part.
+3. **LATAM is underserved**. Verkada and Rhombus don't sell well there. Hikvision dominates but on price and supply-chain concerns. Blocao has room.
+4. **Demo-quality first impression matters**. A SCADA-style live diagram landing view is worth more than a feature-checklist marketing page.
+
+If any of these turn out wrong, here are the failure modes:
+
+- Sovereignty turns out to not be a strong enough trigger → add cloud-bridge as opt-in to lower friction.
+- Integration moat erodes (someone packages Frigate + Mosquitto + console) → keep moving fast on multi-AI fusion and evidence chain.
+- LATAM doesn't grow → pivot focus to EU SMB.
+- Demo doesn't land → iterate the demo until it does, this is recoverable.
+
+## What success looks like
+
+End of year 1: 5-10 customers running Blocao in production, no major sovereignty incidents, MVP demo converts at >25% of qualified leads.
+
+End of year 3: 200+ sites, recognized as the sovereign alternative in LATAM and parts of EU, transcript product validated, first acquisition interest if we want it.
+
+Above all: the platform that lets a small town, a logistics yard, a marina, or a regional bank branch get serious surveillance without sending their video to a US tech company.