- /
- Decks/
Decks
Every article as a short slide deck, made for presenting the idea to your team, your leadership or a customer. Open one and use the arrow keys, click or swipe. Each slide has its own link, so you can share exactly the one you mean.
PDFs are free for newsletter subscribers. Subscribe, open the library link from the welcome email in this browser, and the PDF buttons work for every deck. Subscribe
AI Agent Security & Governance
One policy engine for every agent: Snowflake enforces Cedar and AuthZEN
How a central Cedar and AuthZEN policy decision point and Snowflake Horizon split the work of agent authorisation, and why it holds when the gateway is bypassed.
Keep Shipping: AI security is how you get to yes
Why AI security speeds up AI projects instead of slowing them down. A short deck for leaders, by Kevin Keller.
Governing Snowflake Cortex AI: access, models, regions and audit
The controls that decide who can use Cortex AI, which models run, where inference happens and how it is audited, with the defaults that catch teams out.
Beyond RBAC: AI agents need purpose-aware access control
Why role-based access can't govern AI agents, which attributes a policy must evaluate, and what the industry still lacks.
Federal Zero Trust for data and AI: one control plane in Snowflake
How Snowflake and Cortex map to NIST 800-207, CISA's Zero Trust Maturity Model and OMB M-22-09, with one policy stack for data and AI.
SMCP: adding security to the Model Context Protocol
A proof of concept that layers authentication, encryption, audit and multi-agent coordination onto MCP without breaking compatibility.
Containing AI agents: defence in depth for production
Why sandboxing generic agents is not enough, and a three-layer architecture of custom agent, container and inference proxy that bounds what agents can do.
The agentic enterprise runs on custom agents, not generic protocols
Why production AI agents need purpose-built design, native tooling and embedded security, and where MCP still fits.
Cortex Proxy: one policy point for every AI coding agent
How a small Rust proxy routes any AI coding agent to Snowflake Cortex, OpenAI, Anthropic or Ollama, and checks prompts against security policy on the way.
Governing AI inference: five security layers for EMEA enterprises
A defence-in-depth architecture for AI agents and inference under the EU AI Act, DORA and NIS2, with Snowflake examples.
Building AI Agents
Rixi: sealed, throwaway sandboxes for AI agents on machines you control
How Rixi ships a project and its exact environment to a remote box and gives AI agents a sealed machine of their own.
Nanocortex Web Agent: knowledge management with graph-powered retrieval
How an open-source Cortex Agent application combines zettelkasten retrieval, a review workflow and admin control for domain Q&A.
TurboQuant KV cache: 128B models on consumer hardware
Why the KV cache, not the weights, limits local LLM context, and how a TurboQuant build of llama.cpp shrinks it up to 5x in one conda install.
TurboQuant: what 6x vector compression means for AI agents
How Google's TurboQuant shrinks embedding vectors to 3 to 4 bits without training, and why that changes the maths of agent memory.
Zettelkasten memory: a knowledge graph that lets AI agents remember
How a memory layer inspired by Luhmann's slip box links, ranks and prunes what AI agents learn, across CrewAI, LangGraph and Claude Code.
Automate Google Workspace with one skill file for AI coding agents
How an open-source skill teaches any AI coding agent to build Google Slides, Sheets, Docs, Forms and Drive workflows from prompts.
Your database is the agent framework: coordinating AI agents with Postgres
How one PostgreSQL instance replaces a message broker, vector store, pub/sub system and graph database for coordinating AI agents.
Postgres Analyst: natural language SQL with a local LLM
An open-source semantic layer that lets people query PostgreSQL in plain English with Ollama, keeping schema and data on their own hardware.
Nanocortex: a blueprint for custom Snowflake Cortex agents
What a 1,600-line Python file teaches about the Cortex Agent API: tools, the agentic loop, self-correction and context, ready to adapt to any domain.
Identity & Access
How OAuth 2.0 and OpenID Connect actually work
The four parties, the three flows you will meet in practice, access versus ID tokens, and the mistakes that cause real breaches.
No service accounts: AI chat that queries Snowflake as the employee
A tested pattern for connecting an enterprise AI chat to Snowflake Cortex Agents so every question runs with the asker's own identity, permissions and audit trail.
From SAML login to Snowflake as the user, without the secondary-role trap
How a SAML app on Entra ID can query Snowflake as each signed-in user, with Snowflake deciding which roles the app may use.
Snowflake managed MCP servers accept external OAuth, with RBAC intact
How external AI agents can authenticate to Snowflake managed MCP servers with your identity provider's tokens, tested end to end.
Cross-account SQL in Snowflake: one account obtains, the other validates
How one Snowflake account runs SQL on another through the SQL API with OAuth, and how OAuth, PATs and key-pairs compare for machine-to-machine access.
How JWT tokens actually work, and how Snowflake uses them
What a JSON Web Token contains, how signing and verification work, and how Snowflake External OAuth relies on them.
Credential vending for AI agents: exchanging a user's JWT for a PAT
How a JWT-to-PAT exchange lets an AI agent act in Snowflake as the actual user, with short-lived, role-scoped, auto-rotating credentials.
Encryption & Data Sovereignty
Computing on data Snowflake can't read: homomorphic encryption
How plain Python UDFs let Snowflake sum, average and score encrypted columns without any key, what it costs and where it stops.
The data, the key, or the switch: customer-held sovereignty controls
Runnable patterns that keep the most sensitive data under the owner's control while Snowflake stays the analytics platform, with costs and limits.
AES-CBC column encryption in Snowflake: four key strategies
Four ways to get an AES key to a Snowflake query, from session variables to cloud KMS and key wrapping, with performance and when to use each.
Data sovereignty first: query on-premise Iceberg from Snowflake
How an outbound-only tunnel lets Snowflake query on-premise Apache Iceberg tables while the data owner keeps the off switch.
Column-level encryption for Snowflake with your own key, via Postgres
How Postgres 17 acts as an on-premise crypto proxy so sensitive Snowflake columns are only ever stored as AES-256 ciphertext.
Decrypt PGP files inside Snowflake and drop the MFT middleware
How a Python UDF or stored procedure decrypts PGP-encrypted partner files on a Snowflake stage, so the private key and the pipeline stay inside the platform.
Query your on-premise data lake from Snowflake through a private tunnel
How an SSH tunnel from Snowflake Container Services reaches data, APIs and models behind your firewall, with no inbound ports opened.
Python Camouflage: format-preserving encryption for Snowflake
How FF3-1 tokenisation keeps PII joinable and format-correct in Snowflake, with your own AES-256 key and tag-based masking policies.
Security Operations
Your log bill is a design choice
Why SIEM and observability costs keep growing, and how an open lakehouse with federated queries cuts them without losing the evidence. A short deck for leaders.
Snowflake XDR: a SOC that runs itself, with a human in charge
How Cortex AI can run detection, triage, investigation and response inside one Snowflake account, with humans gating the risky steps.
Cortex Agents in Microsoft Sentinel: closing the AI blind spot
How to send Cortex Agent traces, token usage and access history into Microsoft Sentinel so a SOC can investigate agents.
Snowflake native webhooks for SIEM: Splunk and Microsoft Sentinel
Three ways to push Snowflake security alerts into Splunk or Sentinel, chosen by how the SIEM authenticates, plus detections and safeguards.
Teach any AI coding agent to run security operations in Snowflake
How an open-source AI skill turns coding agents into security engineers that build detection pipelines with OWASP, MITRE ATT&CK and NIST CSF built in.
Splunk + Snowflake: a hybrid security data lake with federated queries
A better-together architecture: Splunk for real-time detection, Snowflake for every log, years of retention and 70-80% lower spend.
An open observability and security data lake: OpenTelemetry into Snowflake
How a Snowflake OpenTelemetry receiver ingests logs, traces and metrics natively and lands them in Elastic Common Schema Iceberg tables.
Let Snowflake raise security incidents in your SIEM automatically
How a Snowflake alert, a table and a scheduled API call turn a failed password login into a SIEM or XDR incident within minutes, with no external code.
Platforms & Infrastructure
DuckIceLake: any Iceberg client on a DuckLake lakehouse
How an Iceberg v3 REST Catalog proxy lets PyIceberg, DuckDB, Trino and Spark read and write a DuckLake-backed lakehouse.
lakesh: one DuckDB shell for Iceberg REST, DuckLake and AI agents
What lakesh adds on top of DuckDB for daily catalog work: profiles, OAuth2, psql-style commands, scriptable output and a read-only MCP server.
One SoftEther hub connects on-premise, AWS and Azure networks
How a single open-source SoftEther VPN hub routes traffic between an on-premise network, an AWS VPC and an Azure VNet.
Snowflake Container Services: security, networking and tunnelling
How SPCS isolates compute, controls ingress and egress, passes caller identity, and connects to your own infrastructure. Updated November 2026.
Bidirectional SSH tunnels from Snowflake Container Services
How an outbound SSH tunnel from an SPCS container gives two-way access to your DMZ, and how to run it safely with secrets, autossh and nginx.
A governed data lake for under 10 euros a month
DuckLake on Hetzner, managed entirely through Pixi, with row-level access control and local-to-production parity.
Python and conda supply chain security needs layers, not just signing
Why signing alone would not have stopped the LiteLLM backdoor, and how signing, curation, rebuilds and runtime containment fit together.
Robyn Extensions: FastAPI-style features with Rust doing the work
How Rust-backed validation, auth, rate limiting and API docs bring a FastAPI developer experience to the Robyn web framework.
How PrivateLink and private DNS keep traffic off the internet
A visual walkthrough of what PrivateLink creates, how private DNS steers traffic to it, and what to secure and budget for.
SPCS Dev Container: a full Linux IDE running inside Snowflake
How Snowflake Container Services (SPCS) becomes a personal cloud IDE with VS Code, a web terminal, persistent storage and direct Snowflake access.
Running Apache NiFi on Snowflake Container Services
A patched, ready-to-deploy NiFi 2.6.0 for SPCS: the ingress fixes it needs, OAuth instead of passwords, and a sample PostgreSQL to Snowflake CDC flow.
Business & Leadership
Your log bill is a design choice
Why SIEM and observability costs keep growing, and how an open lakehouse with federated queries cuts them without losing the evidence. A short deck for leaders.
A few experts, hundreds of sales engineers
How a small group of specialists can scale security expertise across hundreds of sales engineers with AI and just-in-time enablement. A short deck for leaders and managers.
Use the decks freely in your own meetings, with a credit to kevinkeller.org. Snowflake-related content is a personal view, not official Snowflake guidance.