This repository packages the SiMa.ai Neat SDK as a Docker image for x86_64 and arm64 Linux hosts.
This branch defaults to the official eLxr 3.0.0 release, with an Agate/Trixie ARM64 sysroot and GCC 14. Neat Core is bundled when a compatible package is available. Daily and pre-release platform builds remain available through explicit selectors.
The main user workflow is:
- Install
sima-cli. - Install the SDK container image.
- Start the SDK and optionally pair with a Modalix DevKit.
- On x86 platforms, during setup optionally install Model SDK extension.
- Use VS Code to attach to the running container for an IDE.
Install the latest published SDK and run setup:
sima-cli neat install sdk@developInstall a released SDK version:
sima-cli neat install sdk@{version}The install command pulls the SDK image and then asks whether to pair the SDK with a Modalix DevKit. If you choose to pair, enter the DevKit IP address when prompted.
If pairing asks to enable passwordless sudo for the DevKit user, press Enter
to accept the default Yes, or enter n to skip the change.
To install the SDK with Edgematic Studio enabled, select the alternative
metadata-edgematic-studio.json package metadata:
sima-cli neat install sdk@develop --type edgematic-studioReplace develop with a published version or branch as needed. This variant
runs sima-cli sdk setup --edgematic-studio, which installs Studio and publishes
its port. DevKit pairing remains optional. Use a sima-cli version that supports
--edgematic-studio; default SDK installation leaves Studio disabled.
The variant is generated by tools/prepare_s3_artifacts.sh and also included
alongside the online metadata in offline bundle publications; Studio setup
requires network access.
You can still install a container image resource directly when needed:
sima-cli install ghcr:sima-neat/sdk:latestInstall a branch-specific image:
sima-cli install ghcr:sima-neat/sdk:feature-devkit-syncThe published image contains the toolchain, sysroot, headers, libraries, NEAT runtime components, and helper scripts needed to build software for SiMa.ai platforms.
Run the SDK setup:
sima-cli sdk setup --devkit {devkit ip address}Open the SDK shell:
sima-cli sdk neatCheck the installed SDK status from inside the SDK shell:
neatThe SDK workspace is mounted at /workspace inside the container.
Neat SDK 3.0 and newer containers automatically receive access to the host Docker builder during setup:
sima-cli sdk setup
sima-cli sdk neatThe SDK includes a pinned Docker CLI and Buildx plugin. Before creating a 3.0+
SDK container, sima-cli verifies that the host Docker CLI, daemon, and active
local Unix socket are available, then mounts that socket. It verifies and
bootstraps the builder with the SDK's pinned Buildx plugin after the container
starts. The SDK does not run a nested daemon. Commands issued in the SDK create
sibling containers managed by the host daemon. Docker socket access grants
host-level control; sima-cli displays this security boundary during setup.
Inside the SDK shell, build a Modalix application image with:
docker buildx inspect --bootstrap
docker buildx build --platform linux/arm64 --load -t neat-app:dev .The application should be cross-compiled against the Modalix sysroot before
the image build. Avoid target-architecture RUN commands when building on a
non-ARM64 host unless the selected builder has ARM64 emulation available.
The SDK image also preinstalls a tested, pinned version of the OpenAI Codex CLI. From the SDK shell, run:
codexThe Codex CLI version is selected from the checked-in components.json compatibility
catalog and checksum-verified during the image build. Upgrade through a reviewed
SDK component update and rebuilt image to retain the qualified version selection.
The update remains in the current container. Recreating the container restores the version bundled with the SDK image.
The Neat VS Code extension lives in vscode-extension/. It adds a SiMa Neat activity bar panel and light/dark themes. SDK assembly bundles its runtime files at /opt/sima-neat/vscode-extension. Updated sima-cli setup runs sima-install-neat-extension as the mapped SDK user to copy and register this bundled version in that user's code-server extensions directory, without a prompt or download. Repeated setup preserves other extensions and refreshes Neat when the bundled contents change.
Codex and Claude extensions remain optional downloads during sima-cli setup. Set SDK_SKIP_CODING_AGENT_INSTALL=1 in the host environment to bypass their prompts and installation, even with --all-extensions; Neat still installs automatically. On new code-server images, both setup and Palette installs resolve exact versions through the SDK compatibility catalog; unavailable or unqualified combinations are blocked. This setting does not remove existing extensions or the Codex CLI included in the SDK image.
You can run Docker directly for low-level debugging:
docker pull ghcr.io/sima-neat/sdk:latest
docker run --rm -it --name sdk --privileged \
-p 9900:9900 -p 9999:9999 -p 10000:10000 -p 9000-9079:9000-9079 -p 9100-9179:9100-9179 -p 8081:8081 -p 8554:8554 \
-v "$(pwd):/workspace" -w /workspace -v /dev:/dev --pid=host \
ghcr.io/sima-neat/sdk:latest /bin/bash -lThe build environment is configured automatically when the container starts. To configure it manually:
source /opt/bin/simaai-init-build-env modalixThe default environment selects the SDK's platform compiler. Builds that require the qualified GCC 16 C++ toolchain can select the bundled alternate compiler explicitly:
source /opt/bin/simaai-init-build-env modalix gcc16This preserves SYSROOT=/opt/toolchain/aarch64/modalix for target C headers
and platform libraries while exporting SIMA_GCC16_ROOT, CC, and CXX from
the complete GCC 16.2 installation under
/opt/toolchain/gcc-16.2.0-aarch64-linux-gnu. GCC 16 owns its matching C++
headers, libstdc++, and libgcc; the compiler installation is deliberately
kept outside the platform sysroot. Applications must deploy a compatible
target libstdc++.so.6 runtime or package their required runtime closure.
The browser-based VS Code environment is provided by code-server and starts automatically with the SDK container. Full Neat SDK containers keep the HTTP editor endpoint on port 9999 for trusted reverse proxies such as an AWS ALB, and also expose a local HTTPS endpoint using the SDK certificate mounted at /sdk-cert. Open the credential-free codeUI URL printed by sima-cli sdk setup, then sign in with the generated password shown by the CLI. Browsers may require trusting the local certificate before loading editor webviews.
To start it manually if supervision is disabled:
sima-codeBy default, it serves /workspace, requires password authentication, and runs as the SDK user configured by sima-cli sdk setup. Set CODE_SERVER_PASSWORD before the container starts when launching it without sima-cli; a strong temporary password is generated and printed otherwise. Set CODE_SERVER_CERT and CODE_SERVER_CERT_KEY to override the HTTPS certificate files, or mount /sdk-cert/neat-sdk.pem and /sdk-cert/neat-sdk-key.pem. Set CODE_SERVER_HTTPS_PORT to change the HTTPS listener port. Set CODE_SERVER_SUPERVISED=0 to disable automatic startup.
See the Arthas SDK integration guide for hosted provisioning, authentication, and expiration controls.
For a one-click experience behind an authenticated gateway such as Arthas, set CODE_SERVER_WITHOUT_PASSWORD=1. This explicit setting takes precedence if launch tooling also injects CODE_SERVER_PASSWORD. The gateway must authenticate and authorize every request, proxy WebSocket upgrades, and be the only network path to code-server; do not expose either code-server port directly. The same option may be used for explicitly trusted local development, but must not be used on a remotely reachable endpoint without such a gateway.
Rollout requires a sima-cli release containing code-server support (#293). Publish that CLI release, update the sima-cli.ref pin in deps/manifest.json from v2.1.18, and install the release on SDK hosts before distributing the new SDK image: older CLIs pass legacy environment variables but still show token URLs and use OpenVSCode extension/service paths. The image smoke workflow installs sima-cli from PyPI, so it also needs the updated published release. The updated CLI retains support for older OpenVSCode SDK images. Install Core containing neat status support (#1107) to retrieve the password again with neat or neat --json; the password is supplied through CODE_SERVER_PASSWORD in the SDK container environment.
The SDK launcher disables Workspace Trust with --disable-workspace-trust on every start. Folders opened in this SDK editor have full features enabled without Restricted Mode or a trust prompt, including tasks and extensions that can execute workspace code.
code-server stores its user data under ~/.local/share/code-server by default. Override that location with CODE_SERVER_USER_DATA_DIR or set CODE_SERVER_EXTENSIONS_DIR for extensions. During the migration, the launcher also accepts the former OPENVSCODE_* variables so current sima-cli releases can start a new SDK image; new automation should use the CODE_SERVER_* names.
For diagnostics inside the container:
supervisorctl status code-server
curl -fsS http://127.0.0.1:9999/healthz
tail -n 100 /var/log/supervisor/code-server.err.logThe SDK supports an NFS-based shared workspace with a DevKit. The workspace is shared bi-directionally between the host and the DevKit.
Start the SDK container with the DevKit IP:
sima-cli sdk setup --devkit 10.0.0.244Open a direct SSH shell to the paired DevKit:
dk shellRun an executable or Python application from the DevKit:
dk /workspace/app-binary-or-dot-py-fileCheck the paired DevKit and its reverse network path to Insight:
dk statusIn addition to the SSH and workspace status, this checks that the DevKit can reach Insight over TCP and that Insight receives a UDP metadata probe.
Build and push an ARM64 container image from the SDK, then download and run it on the paired DevKit:
docker buildx build \
--platform linux/arm64 \
--tag "${SIMA_CONTAINER_REGISTRY}/hello-neat:develop" \
--push \
.
dk container deploy hello-neat:develop --detach --name hello-neat --network hostdeploy always downloads the current image before running it. Use
dk container run hello-neat:develop ... to run the image already available on
the DevKit. Put Docker run options after the image name. Use -- before a
container command and its arguments.
SDK setup does not install Docker on the DevKit. If Docker is missing, the
first dk container command offers to install it with the required Modalix
/data storage, nftables, and scoped registry configuration. Use
dk container setup --yes for an explicitly approved noninteractive install.
