ETRI TANGO 2 Opens Its No-Code LLMOps Software
On November 9, 2026, South Korea’s Electronics and Telecommunications Research Institute (ETRI) announced the release of TANGO’s no-code AI development technology. The announcement identifies LLMOps, the generative AI component of TANGO 2, as the code published on GitHub. ETRI continues to maintain the earlier computer-vision-focused TANGO 1. The November news is the institute’s disclosure of the software and its use cases. The public repository predates the announcement.
What is TANGO 2?
TANGO stands for Target Aware No-code neural network Generation and Operation framework. ETRI describes it as an integrated way for domain specialists to build, optimize and deploy models. Its public repository groups platform services, target-device runtime components and domain demonstrations. It presents the project as a proof of concept for software-defined industry scenarios, so adoption depends on evaluating individual components and how they fit the target environment.
What changed from TANGO 1?
| Area | Earlier TANGO workflow | TANGO 2 direction | Developer impact |
|---|---|---|---|
| Model focus | ETRI describes TANGO 1 as centered on computer vision | TANGO 2 adds a generative AI operations platform | Evaluate workflow design, evaluation and deployment together |
| Software layout | Automated neural network development and deployment | The repository separates LLMOps, runtime components and field demonstrations | Review the specific module your project will use |
| Target deployment | Models could be prepared for target devices | Runtime components cover optimization, inference and remote management | Test performance on the actual hardware |
| Integration | A model deployment was one part of an application | LLMOps components expose small REST interfaces within the platform | Verify contracts before exposing a service to external callers |
| Release status | TANGO 1 continues to be maintained | ETRI announced the LLMOps disclosure on November 9 | Keep separate baselines and evaluate the new workflow on its own terms |
How is the repository organized?
The TANGO2 repository describes three main areas. LLMOps contains the service platform, web interface, deployment infrastructure and shared libraries. The repository says individual application components expose small REST APIs for platform orchestration. Runtime_System contains target-side pieces for work such as optimization, inference and remote node management. Field_Test contains domain examples. These areas illustrate different responsibilities. An example under Field_Test is not evidence that an arbitrary production workload can be deployed as-is.
The README points to its Kubernetes and Helm installation material under LLMOps/devops. Start there and record the branch or commit you reviewed. Check the required cluster access, container images, storage and any accelerator settings against that revision. The repository describes different responsibilities across its modules. A universal OS, programming-language or model-support matrix was not established from the material reviewed here. Select a module first and use its manifests and component documentation to identify runtime prerequisites. No installation was performed for this report.
What does no-code change for developers?
No-code can move configuration and model workflow assembly into a visual or guided interface. It still leaves data selection, model choice, permissions, runtime constraints and evaluation to people operating the system. Treat those decisions as reviewable configuration. Record the evaluation dataset and what would cause a deployment to be rejected. For example, an industrial inspection application might flag a sample for review when confidence is low, with an operator deciding what to do next.
A .NET team could keep business rules in its ASP.NET Core API and connect to a separately deployed model component through a controlled integration. Start with a representative read-only task. Specify what the request includes, how long the integration may wait, and which result fields are safe to display. Test unavailable dependencies, invalid responses and load on the actual target device. Version the integration contract so a component upgrade does not silently change production behavior.
What should teams check before adopting TANGO 2?
The repository agreement identifies GNU GPL v3.0 or later as the project’s baseline and separately describes ETRI commercial terms. Read both sections and clarify which terms apply to your intended use with ETRI before adopting or distributing the software. Review model weights, datasets, images and third-party dependencies under their own terms. The maintainer’s repository and ETRI’s announcement establish the public code release. They do not by themselves provide a support commitment or settle every component’s deployment rights.
Review who can submit datasets, create workflows, access model credentials and approve deployment. Use a nonproduction namespace and non-sensitive samples for initial evaluation. Audit service-to-service identity and network permissions. Avoid treating container isolation as the whole trust boundary. A long-running evaluation may also use more GPU or storage capacity than a small example suggests, so measure resource use on your own workloads.
The ASP.NET PB API versioning guide explains how to keep client contracts predictable when one side changes.
How can an engineering team evaluate a visual workflow?
Start with one narrow task and record its inputs, expected outputs and failure conditions. A document-triage workflow might accept a synthetic maintenance note, classify its category and leave anything ambiguous for manual review. Track which version of the model and configuration produced each result. A visual editor should make a workflow easier to inspect, but the team still needs a reproducible representation of what it deployed.
Compare the model-facing workflow with the operational boundary. The application may have a five-second response budget, while a batch inference job can run later and report progress. Decide which calls belong on the request path and which belong behind a queue. Exercise a slow component, an unavailable model and a target-device deployment with limited memory. These checks tell you whether the integration can fit your service, even if the demonstration itself runs successfully.
For a team that has already adopted TANGO 1, keep existing evaluations and deployment manifests available during the comparison. Check whether the selected TANGO 2 component actually needs a new model, infrastructure, permissions or data pipeline. The repository describes a follow-up project, so an in-place upgrade should not be assumed.
Frequently asked questions
Is TANGO 2 the same as TANGO 1?
ETRI presents TANGO 2 as a follow-up with an LLMOps component for generative AI. It continues to maintain the earlier computer-vision-focused project.
Does TANGO 2 include a hosted API or a .NET SDK?
The project documents component REST interfaces and a deployment layout. An officially supported public managed API or .NET SDK was not confirmed for this report. Verify the interface on the component you deploy.
Can I install it from a single local command?
Use the repository’s version-specific installation material. The project describes a Kubernetes and Helm deployment path for its LLMOps platform. A universal one-command local installation was not verified.
Comments
Post a Comment