Augmented S3
for the AI era.
One bucket where every object is searchable by meaning, transformable by any AI pipeline, and analyzable across millions of siblings.
AI Agents are ready.
Infrastructure isn’t.
The gap isn’t model capability. It’s the data plumbing underneath every agent.
- Knowledge in silos. Agents can't retrieve.
- Every upload re-stitches the same five pipelines.
- Cross-object insight needs a whole second stack.
A $4.4T opportunity.
Here’s the slice we go after.
McKinsey puts AI’s annual productivity opportunity at $4.4T. AI infra spend is the fastest-growing segment in cloud — and the wedge for an Augmented S3 that does RAG, transform, and analyze natively is exactly where K3 lives.
Worldwide AI infrastructure spend
Gartner forecasts total worldwide AI spending at $2.5T in 2026, with $1.37T concentrated in AI infrastructure — compute, storage, networking, and platform services purpose-built for AI workloads.
Vector DB + RAG infrastructure
K3's wedge sits at the intersection of vector databases ($3.2B in 2026) and RAG infrastructure ($3.3B in 2026) — the two segments K3 collapses into one S3-compatible bucket. Combined growth rate: 25%+ CAGR through 2030.
Sovereign deployments — UK, MEA, EU
Bottom-up: 5 paying customers + 5 named enterprise POCs (Crédit Agricole, PepsiCo, NBE, Rubix) + 40+ early-access pipeline, against a $13.5B MEA sovereign cloud market and €100B European sovereign cloud opportunity by 2031.
Four products.
One platform.
Storage, compute, orchestration, and retrieval — designed to compose into one bucket. Built on the same control plane, the same auth, and the same multi-tenant isolation.
Augmented S3.
An S3-compatible bucket that searches, transforms, and analyzes every object — RAG, AI pipelines on every upload, and cross-bucket analysis, all from one bucket.
Serverless functions for AI agents.
A FaaS runtime designed so AI agents can author, test, and call their own tools. Python and Rust, MCP-enabled, ~200ms warm cold starts, massively parallel by default.
Workflows that compile.
A purpose-built DSL for LLM-driven workflows — 15 composable primitives, full type checking, resumable threads, first-class human-in-the-loop. The language K3 transforms run on.
Managed vector database.
Drop-in pymilvus compatibility, instant serverless pools, automated backups, multi-tenant isolation. The vector index every K3 retrieval call hits.
K3 — three jobs,
one bucket.
K3 is Augmented S3 — vector, transform, analyze, and an embedded warehouse pipelines offload into (HTAP soon). Open it in a spreadsheet or ask the built-in support agent.
Find what's relevant, instantly.
Hybrid semantic search across your bucket — dense vectors, BM25, multimodal queries (text, image, audio, video), and reranking. The full RAG stack, just sitting there waiting for a query.
Run an AI pipeline on every upload.
Trigger a Scriptum pipeline on every object that matches your rules — summarize, redact, caption, transcribe, classify. Outputs land back in the same bucket under a derived prefix like `summaries/`.
Make sense of the whole bucket.
Run Scriptum pipelines across many objects at once — outputs offload into K3's embedded warehouse (HTAP soon), queryable in SQL or the Excel interface alongside the bucket itself.
Where K3 saves time today.
Six industries. Same bucket. Same fifteen pipelines, all shipping day one.
Hybrid search catches semantic clause variants keyword review misses entirely.
Every answer carries a citation chain: policy → regulation → guidance.
Auto-PII redaction and fraud signal scoring run inline at ingest, not weeks later.
Defect patterns auto-cluster across years of FMEA reports — root causes that no single engineer would have spotted.
Multi-tenant + sovereign by default — your enterprise customers get data residency for free.
Foreign-language source material auto-translated and indexed alongside English; multi-classification handling out of one bucket.
K3 doesn’t sit alone.
The same control plane runs three more products that compose with it — compute for the agents that build on it, the language they orchestrate in, and the vector index every retrieval call hits.
Serverless functions for AI agents.
A FaaS runtime designed so AI agents can author, test, and call their own tools — Python and Rust, ~200ms warm cold starts, MCP-enabled, massively parallel by default.
LLM workflows that compile.
A purpose-built DSL with 15 composable primitives, full compile-time type checking, resumable threads, and first-class human-in-the-loop. The language K3 uses for every Transform and Analyze pipeline.
Managed vector database.
Drop-in pymilvus compatibility, instant serverless pools, automated backups, multi-tenant isolation. Powered by Milvus 2.6 and exposed inside every K3 bucket as the underlying vector index.
One bucket, where everyone else needs four boxes.
The capabilities investors care about, scored across the patterns customers actually compare us to.
| Capability | Dodil K3 + Ignite + Scriptum + VBase | AWS S3 + Pinecone + Lambda Hyperscaler patchwork | MongoDB Atlas Vector Managed DB vendor | Vercel + LangChain + Supabase AI app stack | Self-built (S3 + Milvus + glue) DIY |
|---|---|---|---|---|---|
S3-compatible storage Drop-in for any existing S3 SDK | |||||
Hybrid semantic search Dense + BM25 + reranking, in the bucket | |||||
AI transform on every upload Summarize, redact, caption, classify, extract | |||||
Cross-bucket analyze pipelines Anomaly detection, batch scoring, datasets | |||||
Workflow language with type checking Compile-time safety for LLM workflows | |||||
Multi-tenant by design Org isolation across storage, search, processing, credentials | |||||
Sovereign / on-prem deploy Run the full stack in your own datacenter |
Our Cloud
Sovereign infrastructure for AI workloads.
Your data plane runs in your datacenter. Our control plane is in London — expanding to the Middle East and Asia.
Run K3 on our cloud.
Or run it in yours.
K3 is the same Augmented S3 either way. Drop the data plane into your own datacenter, keep every object in-country, and we manage the upgrades, observability, and pipelines. The control plane sits in London today; the data plane sits wherever your compliance officer wants it.
- Run frontier models and agent workloads within border.
- Keep datasets and indexes inside your datacenter (data plane).
- Cloud-grade UX with managed upgrades and observability.
- Control plane hosted in London today; expanding to more regions soon.
Traction
Demand is signed.
The bottleneck is capacity.
Five paying customers. Forty-plus on the early-access list. Five enterprise POCs across four countries — all on the sovereign K3 deployment path.
Enterprise POCs
Compliance research + auditor-grade citation chains across policies, regulations, and guidance — running on K3 in their own jurisdiction.
Defect intelligence and supply-chain knowledge retrieval across decades of QA reports — local data plane, no upstream egress.
AI-powered customer service and fraud signal scoring — sovereign deployment for the National Bank of Egypt.
Multi-tenant K3 deployments across Rubix's regional client base — one platform, many sovereign data planes.
Veterans of the stacks we’re rebuilding.
Investment Opportunity
More demand than we can fulfill.
Use of Funds
- Sovereign GPU footprint to close the next 20 enterprise POCs
- Multi-region data plane expansion (UK → MEA → EU)
- Storage & networking buildout for K3 production tenants
- Platform reliability and 24/7 sovereign-tier SLAs
- Developer relations & community
- Enterprise sales & partnerships
- Regional market expansion (MEA)
- Brand awareness & content
The Challenge: Demand Exceeds Capacity
We have more demand than we can fulfill. 5 Enterprise POCs, 40+ early access signups, and a growing waitlist. This round is about scaling infrastructure to capture the opportunity — not finding product-market fit.
Thank You
Building Augmented S3 for the next decade of AI.
