Federico Cargnelutti @fedecarg

I write about product development and AI at https://t.co/eK0lS7gqY1 London, England Joined December 2008

Tweets

431
Followers

3K
Following

2K
Likes

132

Federico Cargnelutti @fedecarg

3 days ago

Mia now supports conversational prompts. Before, you had to use structured prompts like: Save this PDF in Mia(workspace: Docs, folder: Research). Now you can just write: "Mia, save this PDF in the Docs workspace, inside the Research folder"

0 0 0 54 0

View Details

Federico Cargnelutti @fedecarg

3 days ago

After feedback from PMs and engineers, Mianotes is evolving from a project knowledge organiser into a bridge between product notes, engineering-ready specs, and agent-accessible context. You'll be able to create and share SKILLs, then apply them directly to your notes. v1.0.0 WIP

1 0 0 71 0

View Details

Federico Cargnelutti @fedecarg

5 days ago

@twtayaan @log_npierce Before, generating code was hard, and maintaining it was hard too. What makes you think AI won't tackle the second problem as well? Why assume devs will still be the main maintainers of code 6 to 12 months from now?

0 0 0 29 0

View Details

Federico Cargnelutti @fedecarg

6 days ago

Products can reduce setup documentation by giving developers a short-lived install command that performs the local setup for them. For example: Want to connect Claude Code to our API or MCP? Go to Integrations > Claude > Generate install command. Done. No docs required.

0 0 0 64 0

View Details

Federico Cargnelutti @fedecarg

7 days ago

Why: SBCs often run inside enclosures, on shelves, or as always-on devices. Once deployed, you need a simple way to see if the system is healthy without SSHing into the box every time. It give you quick health check for memory, CPU, storage, and uptime. github.com/senseibox-code…

0 0 0 52 0

View Details

Federico Cargnelutti @fedecarg

7 days ago

I'm also open sourcing a lightweight Linux system telemetry and KPI dashboard for ARM-based SBCs. It's useful for tracking memory, CPU, storage, and uptime on small Linux devices. It's written in Python, exposes REST APIs, and uses WebSockets to update data in real time.

1 0 0 68 0

View Details

Federico Cargnelutti @fedecarg

7 days ago

github.com/senseibox-code…

0 0 0 31 0

View Details

Federico Cargnelutti @fedecarg

7 days ago

Why: The UK NCSC doesn't allow IoT products to ship with factory default passwords, so devices need a secure first-run setup flow. This is similar to the onboarding software you find in proprietary hardware like the Philips Hue Bridge and many always-on bridge devices.

1 0 0 78 0

View Details

Federico Cargnelutti @fedecarg

7 days ago

I'm open sourcing a useful app I built for IoT devices: 1. Senseibox Onboarding: A Linux port of the Arduino App Lab onboarding app. It runs on boot, directly on the IoT device, so users can connect to Wi-Fi and set up the device in order to meet UK security requirements.

1 0 0 83 0

View Details

Federico Cargnelutti @fedecarg

a week ago

I'm releasing the first public version of Mianotes today: → Website: mianotes.com → Docs: mianotes.com/docs/ → GitHub: github.com/Mianotes If you try it, I'd love to know what you'd expect from a tool like this tally.so/r/xXvQbk @msdev @ollama

0 0 0 151 0

View Details

Federico Cargnelutti @fedecarg

3 weeks ago

Transcripts now appear in real time while Mia processes long video or audio. It down-converts long audio to a a mono MP3, slices it into small timed chunks, transcribes each chunk independently, and appends text as each part completes. Works with podcasts and YouTube videos.

0 0 0 88 0

View Details

Federico Cargnelutti @fedecarg

3 weeks ago

The job console is Mia's black box recorder for devs. You can see what the agent is doing, why jobs are failing, and share useful logs to get issues fixed quickly.

0 0 0 66 0

View Details

Federico Cargnelutti @fedecarg

3 weeks ago

@youyeetoo For 12GB with no NPU or powerful GPU, £550 feels a bit expensive to run OpenClaw. You can probably fit 1B to 3B models in there, but still. Add £200 more and you can get a Mac mini with 16GB, fast Neural Engine, and massive memory bandwidth.

0 0 0 65 0

View Details

Federico Cargnelutti @fedecarg

3 weeks ago

@realhamaddada The problem with RKLLM is that it doesn't support vision models. This AIBOX ships with DeepSeek's smallest model (R1-Distill-Qwen 1.5B). Much bigger than YOLOv8, but a different type of model. For YOLO, I'd look at the Dragon Q6A instead. It's a beauty. radxa.com/products/drago…

0 0 1 73 0

View Details

Federico Cargnelutti @fedecarg

4 weeks ago

The AIBOX with Rockchip RK3576 is an interesting computer. It ships with DeepSeek running completely offline. My guess is they used Rockchip's native RKLLM toolkit to install a quantised DeepSeek model small enough to fit into the device’s 8GB of RAM.

1 0 1 126 0

View Details

Federico Cargnelutti @fedecarg

4 weeks ago

@simonw Hey Simon, I know the repo hasn't been updated in a while, but just so you know, Google shows "Access blocked: Google Calendar auth demo has not completed the Google verification process" when running google-calendar-to-sqlite auth.

0 0 0 31 0

View Details

Federico Cargnelutti @fedecarg

a month ago

👉 MCP: Agent-to-tool (Anthropic) 👉 A2A: Agent-to-agent (Google)

0 0 0 47 0

View Details

Federico Cargnelutti @fedecarg

a month ago

Soon your agents will start communicating and collaborating with others. The evolution of work is inevitable. - Legacy: One employee + personal computer - Current: One employee + personal computer + multiple agents - Future: One employee + shared computer + agent-to-agent (A2A)

1 0 1 65 0

View Details

Federico Cargnelutti @fedecarg

a month ago

@lennysan Conductor app 🔥

0 0 1 31 0

View Details

Federico Cargnelutti @fedecarg

a month ago

Should AI-first companies invest more in "AI expert personas" as decision interfaces? For ex: finance, legal, customer, innovator personas. I like this quote: "Skills are tools. Personas are lenses. Agents need both."

0 0 0 43 0

View Details

Federico Cargnelutti @fedecarg

a month ago

Last year I wrote "Turning an LLM prompt into a multi-agent discussion", with multiple AI personas working as a decentralised reasoning engine. Today I saw a great example of this from @takechanman1228 github.com/takechanman122…