SiMa Neat Apps provides runnable C++ and Python examples for detection, segmentation, pose estimation, streaming, tracking, benchmarking, and GenAI. Install Apps to run the packaged examples. Clone this repository only when modifying or building their source.
Run the installation on the Modalix DevKit. If the DevKit has no internet access, run it under /workspace in a Neat SDK container paired with the DevKit. Pairing normally shares /workspace with the DevKit over NFS, so the bundle appears at the same path there. Run the examples on the DevKit.
sima-cli neat install apps also offers to install the compatible Neat Library and Insight. It reads what is installed from neat --json --offline and lists the installed versions under the compatible Neat Library version:
- It skips a Neat Library that matches the compatible version. Otherwise it asks before installing it. Press Enter, or give no answer as in a script, to install it. Answer
nto keep the installed Neat Library. - From an SDK container paired with a DevKit, it also checks the DevKit's Neat Library over SSH and can install it on both. If the SDK is not paired or the DevKit is unreachable, it warns that the DevKit will not get Neat Library and asks whether to install it in the SDK only. To pair, run
sima-cli sdk setup --devkit <devkit-ip>on the host. - It asks before installing Insight when none is installed. It also asks before updating an installed Insight when
neat --json --channel mainreports a different version, because the update installs the latestmain. Answernto keep it. Without network access, it keeps the installed Insight.
To keep the installed Neat Library and Insight without being asked, run the install as NEAT_APPS_SKIP_DEPENDENCIES=1 sima-cli neat install apps. Examples built for a different Neat Library may fail with it. See Dependency Selection.
Install the latest Apps runtime from main and enter the installed bundle:
sima-cli neat install apps
cd prebuilt-appsUse -d <install-root> to install prebuilt-apps/ under another directory. Run the remaining commands from the installed prebuilt-apps/ directory.
Run the same command with the same installation root to upgrade Apps. Before replacing an existing installation, the installer asks for confirmation, carries models/ into the new bundle, and retains the remaining previous files under .prebuilt-apps-backup/. If that backup already exists, the installer warns before replacing it.
Each example README identifies its model, inputs, configuration, and exact run commands.
C++ applications use the packaged executable:
./examples/<category>/<example>/src/cpp/pre-built/<binary> \
--config examples/<category>/<example>/src/common/config.yamlPython applications use their packaged entrypoint:
source ~/pyneat/bin/activate
pip install -r examples/<category>/<example>/src/python/requirements.txt
python3 examples/<category>/<example>/src/python/main.py \
--config examples/<category>/<example>/src/common/config.yamlSome Python applications provide setup.sh and run.sh wrappers instead.
Follow the selected example README.
Browse the Neat Apps Portal or the examples directory to choose an example.
Models are user-managed and are not included in the Apps package. Store them
under models/, or set model.path to another readable package. Each
example README lists its default and additional supported models with the
corresponding sima-cli or example-specific installation command.
Check the installed platform version to confirm the target is supported:
cat /etc/buildinfoModel packages are versioned separately from the platform. They come from the
Model Zoo release, which can differ from the installed platform version.
Model Zoo downloads need your SiMa Developer Portal login: run sima-cli login once, or follow the device-login prompt the first download prints.
Before downloading a model, set the Model Zoo version:
export MODELZOO_VERSION="3.0.0"Runtime sample data is included under assets/datasets/.
| Path | Purpose |
|---|---|
examples/<category>/<example>/README.md |
Example setup and run instructions |
examples/<category>/<example>/src/cpp/pre-built/ |
Packaged C++ executables |
examples/<category>/<example>/src/cpp/ |
C++ implementation reference |
examples/<category>/<example>/src/python/ |
Python implementation and dependencies |
examples/<category>/<example>/src/common/ |
Shared runtime configuration and data |
assets/datasets/ |
Shipped runtime sample data |
models/ |
User-managed model packages |
deps/neat-apps.json |
Installed Apps ref and exact commit |
deps/neat-core.json |
Core artifact selected for this Apps build |
The installed bundle does not contain CMake files or tests. Clone the repository to compile source or run the Apps test suite.
Insight can host videos as streaming sources. Install videos directly from the Insight catalog or through YouTube support, start the required streams in the Insight Web UI, and copy their source URLs into the example configuration. Each example README identifies the required transport, including HTTP MJPEG where supported. Ensure each URL is reachable from the target.
Clone and build Apps inside the Neat Development Environment:
git clone https://github.com/sima-neat/apps.git
cd apps
./build.sh --cleanbuild.sh builds the repository; it does not run tests. Contributor,
scaffolding, build, and test instructions live in
CONTRIBUTING.md.
To fetch one standalone example without cloning the complete repository:
curl -fsSL https://raw.githubusercontent.com/sima-neat/apps/main/scripts/get-example.sh | bash -s -- <example>| Path | Purpose |
|---|---|
examples/ |
Application examples organized by category |
assets/datasets/ |
Runtime datasets shipped with Apps |
assets/datasets-test/ |
Repository-only test fixtures |
models/ |
Ignored user-managed model packages |
tests/ |
Test runner and repository test support |
deps/manifest.json |
Source dependency and platform declaration |
portal/ |
Examples portal source |
deps/manifest.json selects Core and records the SDK channel, the platform version, and the Model Zoo version used for model downloads. Development uses the snap policy, while a release-preparation branch can select an exact Core artifact. Vulcan records the resolved Core target under package metadata and the Apps installer uses that same target. Insight is not pinned in the Apps manifest.
With snap, an Apps tag build selects the same Core tag: Apps v0.6.0 uses
Core v0.6.0:latest, resolved to its published revision and recorded in the
package. Local checkouts exactly at a tag use the same rule. Missing tagged
Core publications fail the build without falling back to develop. Untagged
branch builds retain their existing branch selection and fallback behavior;
explicit manifest pins and --neat-core-version overrides remain authoritative.
Builds inherit the sysroot requirement from Core's bundled Scheduler. Official releases require a matching cached SDK receipt, package inventory and package identity, without overlays; they never run sysroot update. Development pins retain the update path. CI uses the same Scheduler policy with the shared compatibility resolver to select and verify DevKit firmware before installation, then shards tests using its resolved labels.
Use GitHub issues for bug reports and feature requests. Include the example, command, inputs, environment, and logs required to reproduce the problem.
For direct help, contact support@sima.ai.