memprobe 0.1.0

Firmware memory budgets and size-regression checks for CI. Your binary stays local.

memprobe

Firmware memory budgets and size-regression checks for CI, from the command line.

memprobe reads the section and symbol table out of your ELF locally and sends only that metadata to the memprobe API, which returns the analysis. Your binary never leaves your machine — only the sizes and symbol names it contains are sent, the same information readelf and nm print.

pip install memprobe

Quick start

1. Create an API key at https://memprobe.dev (Account → API keys).

2. Point the CLI at it:

   memprobe config set --key mp_live_xxxxxxxx
   

3. Check a build against budgets:

   memprobe init            # writes a memprobe.toml with flash/ram budgets
   memprobe check build/firmware.elf
   

   check exits non-zero when a budget is exceeded, so it gates a CI job.

Commands

Command	What it does
memprobe analyze <elf>	Size summary: flash/ram totals, biggest sections and symbols.
memprobe check <elf>	Fail (exit 1) if a budget in memprobe.toml is exceeded. The CI gate.
memprobe diff <old> <new>	Size change between two builds, with per-symbol deltas. --format markdown for PR comments.
memprobe init	Scaffold memprobe.toml with flash/ram budgets.
memprobe account	Show your plan and this month's usage.
memprobe config set --key <key> [--server <url>]	Store your API key (in ~/.memprobe/config.json).
memprobe config show	Show the current key (masked) and server.

MEMPROBE_API_KEY and MEMPROBE_SERVER environment variables override the stored config, which is convenient in CI.

CI

Because memprobe check exits non-zero when a budget is exceeded, it works as a gate in any CI system. Run it as a build step with MEMPROBE_API_KEY set as a secret:

pip install memprobe
memprobe check build/firmware.elf

What runs where

	Local (this tool)	memprobe API
Reads your ELF	yes	never sees the binary
Extracts sections/symbols	yes (via pyelftools)	—
Budget / diff / bloat analysis	—	yes

This package contains no analysis logic and no proprietary code — it's a thin, open-source client. The deeper analysis (call graph, dead-code, stack usage, source attribution) lives in the web app at https://memprobe.dev.

License

MIT