A note about this build log entry: I published this after the week it covers. DevNitro reconstructed a draft from work-related entries in my journal, Git commit history, and time tracking. I reviewed, edited, and approved the final post. The story reflects what I knew and built at the time.
From the Backlog to the Build Log
Week of March 20 - March 26, 2026
What I Worked On
On Thursday I had to replace a dead car battery. I pulled it, drove to the shop, waited, grabbed food, and drove home. The whole time, DevNitro was working through issues from my phone. It even ran a database deployment while I sat in a parking lot. That was the week.
The feature behind it is a CLI relay. The agent runs on your machine, where your code and your tools live, but a relay protocol over SignalR lets the web UI, or a phone browser, talk to that running agent in real time. You get a live processing heartbeat so you can see it is alive, real-time LLM token usage so you can watch the cost as it runs, and conversation affinity routing so a chat always reaches the same local agent instance instead of a random one. You start a conversation from the web, it runs on your machine, and the results stream back.
The rest of the week was about giving that agent good raw material. I stood up a private NuGet feed with credentials that auto-provision and cache, wired through Docker build secrets so nothing leaks into an image. And I built a blueprint catalog: a policy system, toolchain checks, rich metadata for each blueprint like engine, tech stack, and requirements, a list_blueprints tool so the agent can discover what is available, and an admin UI to manage them. This is the next layer on the scaffolding work from last fall and the distribution work from earlier this month.
What the AI Did Well
The blueprint catalog is exactly the kind of broad, structured feature the agent builds well: an API, a client, a local cache, policy checks, and admin CRUD all following the same shape. It also built an admin financial dashboard with real SaaS metrics, and added a global AI provider and model catalog so I can manage which models are available. That last one I deliberately hid in the web UI for now, because I want provider and model choice to be restrict-able by company and team before I expose it, and shipping it half-governed would be worse than not shipping it.
What Needed Correction
The relay did not work on the first try. It connected, but the language model responses would not stream back, so I had a live pipe with nothing flowing through it. Getting the streaming right took a few iterations.
The more important correction was security. The moment you let a client talk to a server-hosted agent, you have to assume someone will hand-roll a malicious client and try to reach conversations that are not theirs. So I spent real time closing that: enforcing conversation ownership on every access, scoping sessions and conversations to a project, and restricting regular members from seeing all of a company's people. A relay is a door. You do not ship a door without a lock.
I also got honest feedback from an early tester using the Copilot path. He got a lot done fast, but he was having a hard time seeing the full value. That stung, and it was useful. It told me the web chat and the VS extension experience had to get a lot better, which is a lot of why this week bent toward the relay and onboarding.
What Shipped
DevNitro has a CLI relay over SignalR that lets the web UI and a phone drive the local agent in real time, with a heartbeat, live token usage, and conversation affinity routing. There is a private NuGet feed with auto-provisioned, Docker-secret-backed credentials. There is a blueprint catalog with policy, toolchain checks, metadata, a list_blueprints tool, and admin management. Onboarding got a VS extension wizard, I got a financial dashboard, and conversation access is locked down by ownership and project scope.
The Numbers
| Category | Hours |
|---|---|
| Product / Features | 33h 49m |
| Testing | 10h 4m |
| Business | 8h 11m |
| Content | 3h 38m |
| Community | 1h 10m |
| Total | 56h 52m |
The eight hours in the business bucket is onboarding, which is fitting. This was a week about the distance between my machine and yours, and onboarding is the last mile of that distance.
Watching the agent close issues from my phone in a parking lot was the first time DevNitro felt less like a tool I run and more like a teammate I check in on. That is the feeling I am chasing. This week I got a real taste of it.
Curious how this site was built?
Watch the 5-minute walkthrough at devnitro.com →
0 Comments
No comments yet. Be the first to share your thoughts!
Leave a comment