Skip to main content
An open agent with a layer of curated IoT expertise on top. The layer is what makes the difference on real silicon, and it comes in two forms. Knowledge bits teach the agent something. Tool bits give it something to run.

Where the agent sits, and what it loads

You, in VS CodeYou approve every step.
The agentPlans, edits, runs, and reads the results back, on the model you chose.
Expert knowledge, loaded for this taskWhat a symptom means, and the tools that capture, decode and scan.
Your toolchainThe vendor SDKs and probes already on your machine.
Your boardThe real one, on your desk.
And back again. Logs, bus traces, radio captures and fault dumps return to the agent. That return path is the whole difference. A tool that only reads your code never sees any of it.
Everything except the expertise layer is yours: your editor, your keys, your model, your toolchain, your board.

Two kinds of bit

Both are scoped narrowly, versioned, credited to the engineer who wrote them, and loaded only when a task needs them.

How knowledge is scoped

A task pulls only what it needs, widest to narrowest: always on, then the silicon family, then the SDK, then the protocol, then the board, then the product. A new silicon family is ordinary work rather than a special case, so the roadmap can add one without touching the agent.

What the tools cover

Which tool answered a question is named in your session at the moment it runs. Every tool bundled in the extension is open source under Apache-2.0. The registry also serves Adsum’s own commercial tools, CRA scanning and cellular among them.

Credited, in your session

Load a bit and a credit line appears in the conversation. Click it and you see who curated it and who maintains it.
Adsum IoT Coder credit line reading Knowledge bit, Add Feature Workflow, curated by Omar Morceli, with a provenance card showing who curated and who maintains it

A credit line in a live session, and the provenance it opens.

Every bit records how far it has been tested on a bench, and newer knowledge earns that record as engineers prove it. The credit line is attribution, not a verdict: it names who built the expertise applied to your task, and says nothing about your device or your compliance status.

Why this compounds

The layer gets better when an engineer proves something on a bench, and stays better because the finding is written down and loaded the next time anyone meets that board. A verified board makes the next one cheaper: the SDK and protocol knowledge is already there, so a new board is just the board rung. Add your own expertise.

Written by people

The knowledge behind Adsum is hand-authored or hand-reviewed by senior IoT engineers, not generated by the agent. Generated material can read fluently and still carry a subtle misunderstanding of a protocol stack, which the agent cannot detect in itself. Expert curation is what keeps the bar honest.