BootLoop

The agent acts on the bench. It drives oscilloscopes, logic analyzers, power meters, GDB, JTAG probes and serial monitors, and it can work through a debug probe before a UART, printf or clock is running. An equipment library documents SCPI and PyVISA control for oscilloscopes, power supplies, source measure units, electronic loads, analyzers and generators from Keysight, Rigol, Tektronix, Keithley, Rohde & Schwarz and others, and the BootLoop Test agent can add support for equipment that is not on the list.

That reach stays inside lab equipment and the customer's toolchain. BootLoop does not describe connectors that let the agent act in business software such as a ticketing system or a repository host.

Sourcebootloop.ai/blog/bootloop-memfault-core-dump-ai-firmware-agentread 2026-10-08

Develop runs a build, flash and test loop on its own until the result converges: it compiles against the customer's toolchain, flashes the board, exercises the firmware with test equipment and patches what fails. Test pairs dozens or hundreds of benches to a central control plane, and because tests are written once and run on any capable bench, the same suite moves from bring up to CI to end of line without rewriting.

BootLoop describes one agent loop plus fleet wide test runs. It does not describe several agents working together or a workflow designer.

Sourcebootloop.ai/products/testread 2026-10-08

Develop grounds its work in the customer's own design files. It parses schematics, netlists, datasheets of any length, reference manuals and Altium, KiCad or Cadence PCB projects, and builds a register and pin level model of the specific board, with peripherals and net connections. A new chip needs no supported list: uploading its documentation is enough. On the test side, each bench is described in a Bench Graph that records the equipment and how it connects.

BootLoop does not say how the board model is stored, kept current or searched across projects.

Sourcebootloop.ai/products/developread 2026-10-08

The build, flash and test loop runs without intervention until it converges, and the engineer then reviews the resulting diff once. BootLoop describes no setting that pauses the agent before it edits code, flashes a board or drives an instrument.

Its launch post adds that during a pilot a person validates every line of code, which describes how pilots are run with BootLoop's engineers rather than a control the customer operates in the product.

Sourcebootloop.ai/blog/bootloop-memfault-core-dump-ai-firmware-agentread 2026-10-08

A trust center at trust.bootloop.ai names SOC 2, without stating its type, along with ITAR compliance for the GovCloud deployment, SSPA and infrastructure it calls DoD ready. BootLoop says firmware builds, flashes and runs on the customer's own hardware, that code is not retained and that customer data never trains models.

Who inside a customer's organization can do what is not published. BootLoop describes no single sign on, roles or permissions, and its Test launch post says results are visible to the whole company.

Sourcetrust.bootloop.airead 2026-10-08

A unified Run History collects bench, CI, end of line and field runs in one web view with full version control. Each result keeps its step results with the bench configuration, test code and firmware artifacts, so a team can see what the agent changed and how the hardware responded, and replay the run on any bench. For regulated and functional safety work, results, binaries, test code and bench configuration are kept for seven years or more, and yield, runtime and pass rate roll up across benches.

These are records of what ran. BootLoop does not describe a log of which user or agent took which action.

Sourcebootloop.ai/blog/bootloop-test-launchread 2026-10-08

BootLoop's pages describe what the agent reads at the start of a task, the board's design files and documentation, but not anything it remembers afterward, such as project conventions, earlier fixes or past decisions it consults on the next job. The Run History keeps every result, binary, test and bench configuration, which records what ran rather than what the agent learned. BootLoop publishes no product documentation that would say more.

Sourcebootloop.ai/products/developread 2026-10-08

Customers can run BootLoop in its cloud, as a single tenant deployment, in an ITAR compliant GovCloud or on premises. Whichever they choose, firmware builds, flashes and runs on their own hardware, and any Linux computer can be enrolled as a test bench.

Sourcebootloop.airead 2026-10-08

An equipment library gives SCPI command references, generated from official programming manuals, and PyVISA examples for a wide range of lab instruments. In Test, equipment connects through a visual interface without code, the BootLoop Test agent generates tests and components from the firmware codebase, schematics and bench, and tests ship with their own binaries so a board can be provisioned without a complex bench.

These are building blocks for benches and tests. BootLoop does not offer a catalog of prebuilt agents or test templates to adopt.

Sourcebootloop.ai/equipmentread 2026-10-08

Tests promoted from the agent's on device checks run on every commit in CI and in nightly and regression runs. On the production line, an operator scans each unit to start its end of line test through the BootLoop Test CI API.

BootLoop does not say which CI systems it plugs into or how a commit reaches the API, and the agent itself starts when an engineer gives it a task.

Sourcebootloop.ai/products/testread 2026-10-08

BootLoop names no model behind its agent, and its product pages and posts offer customers no way to choose or bring one. The agent works in C, C++ and Rust across Zephyr, bare metal and embedded Linux, but which AI models do that work is not said.

Sourcebootloop.ai/products/developread 2026-10-08

A BootLoop Test CI API exists: the Test launch post uses it to start end of line runs when a production operator scans a unit, and promoted tests run through CI without being rewritten for a different runner. BootLoop publishes no reference for that API, no SDK and no MCP server, and the command line interface in its product images has no public documentation.

Sourcebootloop.ai/blog/bootloop-test-launchread 2026-10-08

Testing is half of what BootLoop sells. Each successful on device check the agent runs can be saved as a deterministic test, copied into BootLoop Test and promoted to CI, regression, nightly or end of line runs, with pass rates tracked across benches. Engineers can also write tests themselves, and BootLoop promises a first test running on the customer's hardware in under an hour.

The agent also tunes for power: given a datasheet, a power meter and a target current draw, it adjusts peripheral settings, sleep modes, DMA and clocks against live measurements. BootLoop's internal benchmarks put a hardware tested peripheral driver at about 8.5 minutes from the prompt on a well supported Cortex-M target.

Sourcebootloop.ai/blog/bootloop-test-launchread 2026-10-08

BootLoop works through instruments, debuggers, probes and boards on the bench rather than through a screen. Equipment is driven over SCPI and debug interfaces, and no browser or desktop is operated by the agent.

Sourcebootloop.ai/products/developread 2026-10-08