Five early-stage infrastructure projects, each aimed at a different corner of the market: AltSql, VPN Works, Precomputing, Preconfiguration and BareProxy. All five are Alphas or 0.1 releases. All five have browser demos. They’ll be presented at Cloudflare Connect 2026, October 19–21 in San Francisco.
The common thread is a bet on small, inspectable tools in markets that keep drifting toward bigger, opaque ones. Whether that bet pays depends less on the code than on where each market is heading. So, market by market.
AltSql and the edge database market
AltSql is a native hybrid of key-value and SQL. About 15 KB of code on a device keeps its records in flash; the gateway runs SQL over those same records, byte for byte. The pitch is the removal of the translation layer every connected product ends up maintaining for a decade. The How It Works page lays it out, the AltSql modules extend it, and the live demos run it in a browser.
The market read: embedded and edge data is crowded at the bottom (SQLite owns the default) and fragmented at the top, where time-series vendors and cloud IoT platforms compete for the gateway. What’s changing is that devices are expected to decide more on their own, with less bandwidth and less cloud. That favours anything that keeps data usable at the endpoint. The risk is just as clear. Device makers buy slowly, qualify hardware for years, and rarely bet on a 0.1 storage engine. AltSql hasn’t run on a real microcontroller yet; until it does, this is a thesis, not a product.
VPN Works and agent security
VPN Works gives one program its own network path: a coding agent sealed in a Linux namespace, every connection checked and logged. VPN Works Scope applies the same draft-then-review approach to company VPNs. The live demo replays an agent trying to leak a token.
This is the market moving fastest. This week alone, Apple tightened Full Disk Access on macOS citing AI agents, and OpenAI disclosed that its own agent had breached an Australian state government. Agent containment is turning from a niche worry into a budget line. The incumbents are well placed though: zero-trust vendors, endpoint security firms and the agent platforms themselves will all ship their own controls. A small tool wins here only if it’s the one engineers reach for first, the way they reach for a firewall rule before a vendor contract.
Precomputing and the cost of observability
Precomputing keeps dashboard answers ready as data arrives, compiled from a short policy into plain SQLite. The policy language explains the mechanics; the SQL demo shows three hours of API traffic shrinking into a small file.
The market behind it is the bill. Observability and usage data grow with traffic, and pricing tied to ingested volume punishes exactly the companies that grow. That pressure is why log reduction, sampling and pre-aggregation keep attracting buyers. The opening for a SQLite-based approach is the long tail: teams too small for a data platform but too big to keep querying raw events. The hard part is distribution. Databases are adopted bottom-up by developers, and a 0.1 release with no production users has to earn that trust one install at a time. This week’s Supabase deal for Turso, an agent-focused database, shows the appetite for small, embeddable databases is real; it also shows that the winners tend to get absorbed by platforms.
Preconfiguration and agent environments
Preconfiguration builds every coding-agent platform’s setup files from one short spec and proves them on a clean machine. Preconfig Doctor reads failed setup logs and names the cause with rules, not a model. The live demo shows the whole loop.
This is a market that barely existed two years ago. Every cloud coding agent starts on a fresh machine, and every platform describes that machine differently. Fragmentation like this usually ends one of two ways: a standard emerges, or a translation layer becomes the standard. Preconfiguration is betting on the second. The risk is that the platforms converge on a shared format themselves, the way dev containers spread, and the problem quietly disappears. The upside is that while they don’t, every team running more than one agent pays the setup tax.
BareProxy and the reverse proxy market
BareProxy is a small web server and reverse proxy that explains every routing decision, with BareProxy modules for caching, limits and the rest. The demo shows its explain, why and plan commands.
The reverse proxy market is mature and dominated: nginx is everywhere, Caddy won over developers with automatic TLS, Envoy and Traefik own much of the cloud-native layer, and CDNs like Cloudflare absorb the rest at the edge. Nobody needs another proxy. What BareProxy sells is different: a config you can predict before you deploy it. Misrouted traffic after a config change is a familiar outage pattern, and change-safety tooling is where proxies still compete. It’s the youngest of the five, days old, and its numbers are design budgets. Presenting it at a Cloudflare event is either bold or cheeky. Probably both.
The portfolio view
Read as a portfolio, the five split cleanly. Two ride a market that’s accelerating right now: agent security and agent environments. One targets a cost problem that grows with every customer’s traffic. Two enter mature markets on a narrow angle, explainability over features.
The shared weakness is stage. None of the five has production users yet. The shared strength is that each one can be judged in a browser in five minutes, which, for early infrastructure, is most of the first sale.