chumicro-deploy
Push code onto a CircuitPython or MicroPython board, probe its identity, and flash firmware, all from your laptop.
Ship a host directory, a file map you built in memory, or only the modules your entrypoint actually imports. Run the payload from RAM while you iterate, or write it to flash when the code has to survive a reboot. When a deploy fails you get the reason (another app is holding the serial port, the CIRCUITPY drive ejected, the raw REPL is stuck, macOS wedged its FSKit filesystem extension) and the steps that clear it, instead of a raw traceback. Available as a Python API and as a chumicro-deploy CLI.
Part of the ChuMicro family: small, focused Python libraries for microcontrollers and laptops. Browse all workbench tools. This is a workbench tool: it runs on your laptop, not on the board.
Install
pip install chumicro-deploy-experimental
pyserial (CircuitPython transport) and mpremote (MicroPython transport) come along as dependencies. On macOS and Linux you'll also need rsync available on $PATH for flash-mode deploys to the CIRCUITPY USB drive. Install it via brew install rsync (macOS Homebrew if not already shipped), apt-get install rsync (Debian/Ubuntu), dnf install rsync (Fedora), pacman -S rsync (Arch), apk add rsync (Alpine), or zypper install rsync (openSUSE). Native Windows isn't currently supported (raises WindowsNotSupportedError on import); WSL2 works.
Experimental (pre-release) versions and channel switching
Pre-release builds are published automatically when the package version is bumped.
pip install chumicro-deploy-experimental
Quick example
Deploy a one-liner to a connected MicroPython board and read its output back:
from chumicro_deploy import Device, Deployer, FileMapSource
device = Device(
transport="micropython",
address="/dev/cu.usbmodem14101", # see `chumicro-deploy probe --address ...` to identify
deploy_mode="ram", # no flash wear, no CIRCUITPY drive needed
)
source = FileMapSource(
{"/main.py": "import sys\nprint(f'hello from {sys.implementation.name}')"},
entrypoint="/main.py",
)
result = Deployer(device).deploy_diff(source)
print(result.execute_output) # → "hello from micropython"
For a workspace project that already has a devices.yml, swap the Device(...) constructor for chumicro_deploy.config.default.load_devices_yml("devices.yml", device_id="my-board").
What's included
Programmatic API
| Symbol | Description |
|---|---|
Device(transport, address, deploy_mode, ...) |
Configure a target board. Build explicitly or via Device.from_dict(...); or load a registry via load_devices_yml(...) / load_device_registry(...) |
Deployer(device) |
Deploy orchestrator for one board. deploy_diff is its one deploy method |
Deployer.deploy_diff(source, *, clean=True, wipe=False, ...) |
Push a FileSource, run the entrypoint, and delete the on-device files that aren't in the new payload. Returns a DeployResult with success / output / traceback. Clean-slate by default; clean=False reconciles only the entrypoint, state files, and /lib; wipe=True erases the filesystem first |
RecoveringDeployer(deployer, *, prompt=None, max_attempts=3) |
Wraps Deployer with failure classification and a RecoveryPlan. Pass prompt=input for an interactive retry loop; default prompt=None reports once and re-raises. Both CLIs use the interactive form by default (--non-interactive to opt out) |
FileMapSource(files, entrypoint) |
In-memory dict[device_path, bytes] source, for generated payloads or one-off scripts |
DirectorySource(directory, entrypoint, resource_prefix) |
Walk a host directory and ship every file under it |
ImportGraphSource(entrypoint, search_paths, device_entrypoint) |
AST-walk the entrypoint and ship only transitively-imported modules |
probe_device(device) → DeviceInfo |
Identify a connected board (runtime, version, machine string, CPU UID) |
flash_firmware(url, device, reflash_method, ...) |
Download + apply firmware via UF2 (Pi Pico family) or esptool (ESP32 family) |
resolve_firmware_url(board_id, runtime, version) |
Build the canonical Adafruit / micropython.org download URL |
classify_deploy_failure(error) → DeployFailureKind |
Standalone classifier for building your own failure-handling on top of Deployer |
detect_fskit_wedge() → bool |
macOS-only probe for the FSKit / DiskArbitration wedge that can leave CIRCUITPY drives unmountable |
devices.yml writer surface
chumicro_deploy.config.devices_yaml (separate submodule) reads and writes the device registry round-trip, preserving comments and key order, and classifies every field into one of three zones (user-owned / hardware-once / probed-always). This is what chumicro-workspace add-device and friends sit on; consumers building their own onboarding flow against devices.yml use this surface directly.
| Symbol | What it does |
|---|---|
read_devices_yml_template() → str |
The empty-registry devices.yml text, for writing a fresh file before the first device is registered |
load_devices(path) → CommentedMap |
Parse with comments + key order preserved (returns ruamel CommentedMap, distinct from the typed reader at chumicro_deploy.config.default.load_devices) |
dump_devices(data, path) |
Atomic write back via tempfile + rename |
find_device(data, device_id) / list_device_ids(data) |
Read-only lookups against the loaded document |
add_device(data, *, device_id, runtime, address, hardware, ...) |
Append a new entry; raises DeviceAlreadyExistsError on id collision |
remove_device(data, device_id) → CommentedMap |
Drop an entry and null any defaults: key that pointed at it. Returns the removed entry, so the same id can be re-registered. Raises DeviceNotFoundError when there's no match |
update_device_address(data, device_id, new_address) |
Silent refresh; address is in the probed-always zone |
update_device_firmware_version(data, device_id, new_version) |
Silent refresh of the cached firmware_version |
update_device_hardware(data, device_id, *, force=False, **fields) |
Hardware-once zone. Raises HardwareOverwriteError on a value change unless force=True |
rename_device(data, old_id, new_id) / set_runtime_default(data, runtime, device_id) |
Higher-level mutations |
USER_OWNED_FIELDS / PROBED_ALWAYS_FIELDS / HARDWARE_ONCE_FIELDS / HARDWARE_BLOCK_ZONES |
The canonical zone classification, the single source of truth that the typed reader's _KNOWN_KEYS derives from |
CLI subcommands
python -m chumicro_deploy <subcommand> (or just chumicro-deploy-experimental <subcommand> after pip install). The three board-facing subcommands accept --devices-file devices.yml --device <id> instead of --transport + --address for workspace-style invocations.
| Subcommand | What it does |
|---|---|
probe |
Identify a board's runtime / version / machine / UID. --json for machine-readable output |
deploy |
Push a --directory or --file-map to the board and run the --entrypoint. Interactive coaching by default; --non-interactive for CI |
flash-firmware |
Download and apply firmware. --method uf2 or --method esptool, inferred from the URL extension when omitted. On the esptool path the flash is erased first, wiping the CIRCUITPY drive, NVS, and stored WiFi credentials; pass --no-erase to keep user data on an in-place upgrade. --non-interactive fails instead of prompting on bootloader entry |
resolve-firmware-url |
Print the CircuitPython firmware URL for a --board-id + --runtime + --version (no board needed) |
Testing fakes
| Symbol | What it does |
|---|---|
FakeTransport |
Implements TransportProtocol + ExtendedTransportProtocol for unit-testing Deployer orchestration without real hardware |
FakeSerialPort |
Scriptable pyserial substitute for CircuitpythonTransport tests |
FakeTime |
Deterministic clock for Deployer / transport / flash_firmware tests |
Where this fits
Leaf: no upstream ChuMicro deps (uses third-party pyserial and mpremote for transport). Sister of chumicro-repl, substrate for chumicro-workspace and chumicro-pytest-device.
Companion: chumicro-repl
chumicro-repl is the sister workbench tool for opening interactive serial sessions and tailing the friendly REPL after a deploy. Both packages consume the same devices.yml schema (owned here in chumicro_deploy.config.default), so a single workspace file points both at the same boards. Use chumicro_repl.tail(device, seconds) to follow a deploy and fail-fast on a traceback; use chumicro_repl.ReplSession(device) for headless test fixtures over raw REPL.
Examples
| Example | What it shows |
|---|---|
programmatic_deploy.py |
Minimal Deployer + DirectorySource walkthrough |
file_map_deploy.py |
Multi-file payload built in memory via FileMapSource |
import_graph_deploy.py |
ImportGraphSource AST-walk that ships only the modules the entrypoint actually imports |
demo_recovery_hand_holding.py |
Interactive walk through every DeployFailureKind recovery scenario against real hardware |
Contributing
Issues, bug reports, and pull requests are welcome, and so is "I ran it on this board and here's what happened", some of the most useful feedback a hardware project can get. Development happens in the ChuMicro repository, whose contributing guide covers setup and the test workflow.
Docs
📖 Stable docs · Experimental docs
Find this library
- PyPI: chumicro-deploy
- Source: workbench/deploy
License
Metadata
Release files for chumicro-deploy-experimental 0.39.3
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | |
|---|---|---|---|
| chumicro_deploy_experimental-0.39.3.tar.gz | 172.7 kB | Details |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Reset |
|---|---|---|---|---|
| chumicro_deploy_experimental-0.39.3-py3-none-any.whl | Python 3 | none | any | Details |
Total release size: 369.5 kB
Release files / chumicro_deploy_experimental-0.39.3.tar.gz
| Download URL | chumicro_deploy_experimental-0.39.3.tar.gz |
|---|---|
| Size | 172.7 kB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
36948f02fa99d0a2c25041a54ee4de506e41f910f769f47346549e2bc3c53d69
|
|
BLAKE2b-256 checksum How to use checksums |
93ceb172746db068133ec5b525dd1618cd17f30be90da46e53b8e61690e401c4
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/7.0.0 CPython/3.13.14
|
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Aug 9, 2026.
Transparency logRelease files / chumicro_deploy_experimental-0.39.3-py3-none-any.whl
| Download URL | chumicro_deploy_experimental-0.39.3-py3-none-any.whl |
|---|---|
| Size | 196.8 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
f1713f41447309886e1d01406c6a764d37e3f9ae64982dceb08ee2007d2b2488
|
|
BLAKE2b-256 checksum How to use checksums |
18caa563d2c740266e477354a56821150c9c97f82c6248c6d90f9fef47fd243e
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/7.0.0 CPython/3.13.14
|
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Aug 9, 2026.
Transparency log