a[ssistan]t
Tools you talk to that take action.
2 projects →I build AI you can leave running.
I'm Leonardo Cardoso, a senior software engineer in Berlin with over a decade in production codebases across backend, web, and mobile, currently at Trade Republic. a7t is where I publish the agentic systems I build off-hours, the ones I actually keep running every day.
The five
a · _ · ta[rchivis]t
Memory you can trust and recall.
2 projects →a[rchitec]t
Infrastructure for agents and pipelines.
2 projects →a[utopilo]t
Routines that hold the line without you.
2 projects →a[utoagen]t
Agents that plan, act, and report back.
1 project →Projects
open sourceA tasteful AI chat you self-host on one machine and open to your household over your own network. Your local models, passkey sign-in, no cloud.
OpenAI-compatible router that greedy-batches calls so a burst of requests costs one model swap.
Five workflows that pick issues from a project board, implement them, fix their own CI failures, merge green PRs, and keep the board updated.
GitHub Actions + Fastlane that take a clean Xcode project from feature-merged to TestFlight delivered, version bump and App Store metadata included.
A native macOS companion that lets you browse Claude Code sessions, explore diffs, and re-run commands, all from a polished SwiftUI interface.
Articles
long-form, slowly writtena7t.chat: The AI Chat My Whole House Uses
For almost a year my local models did real work for exactly one person: me, through a terminal. The rest of the house had no way in. a7t.chat…
Geofenced Out of the Frontier: Why I Run My Own Models
Two labs geofenced their best models behind US compliance in a single week: OpenAI's GPT-5.6 preview to US partners only, Anthropic's general model shut down and still offline,…
When the Reader Is a Robot: I Tried to Fix Ads for AI Agents, Then Killed the Idea
The agentic web breaks the old content-for-traffic-for-ads handshake. I designed a platform to inject ads into agent-read text and pay humans to look, then pressure-tested it against unit…
The Router Sees Everything: Adding Telemetry to Jano
A model server can only tell the one client it just answered what that response cost. It can't see the queue, the swap thrash, or the throughput across…
The MoE Speedup, Measured: 50 Prompts, Two Local Qwen Models, One Mac Studio
A dense 27B at Q8 runs my local chat at 15 tokens a second. Its 35B Mixture-of-Experts sibling runs the same prompts at 70. I ran 50 through…
The Bottleneck in Enterprise AI Isn't the Model. It's the Context Layer Underneath.
Every team building AI inside a company hits the same wall: the model doesn't know who you are. Why vector RAG hits a ceiling on relational data, what…
Booklets
Architectural deep-dives, one-time price. Lifetime updates.v0.1 · 41 pages
Three-Body Agent: the Playbook
Blueprints for autonomous frontier LLM pipelines that carry a GitHub issue to a merged pull request, with no human in the loop.
Coming soon
Slipway: the Playbook
The iOS release pipeline, written down: GitHub Actions and Fastlane carrying a merge on main all the way to a TestFlight build.
Contact
work with meWant one of these wired into your stack, or a system spec'd from scratch? Send me an email.
Stay in the loop
One email when something ships.Get told when the next booklet ships, plus the free seven prompts behind Three-Body Agent 1-pager.
No spam. One email when something ships. Unsubscribe in one click.