> This page is for version 0.8.0. > For other versions, use one of these documentation indexes: > - Latest (default): https://docs.nvidia.com/nemo-helix/latest/llms.txt > - 0.8.0: https://docs.nvidia.com/nemo-helix/v0.8.0/llms.txt > - 0.7.1: https://docs.nvidia.com/nemo-helix/v0.7.1/llms.txt > - 0.7.0: https://docs.nvidia.com/nemo-helix/v0.7.0/llms.txt > - 0.6.0: https://docs.nvidia.com/nemo-helix/v0.6.0/llms.txt > - 0.5.1: https://docs.nvidia.com/nemo-helix/v0.5.1/llms.txt > - 0.5.0: https://docs.nvidia.com/nemo-helix/v0.5.0/llms.txt > - 0.4.0: https://docs.nvidia.com/nemo-helix/v0.4.0/llms.txt > - 0.3.0: https://docs.nvidia.com/nemo-helix/v0.3.0/llms.txt > - 0.2.0: https://docs.nvidia.com/nemo-helix/v0.2.0/llms.txt > For clean Markdown of any page, append .md to the page URL. > For a complete documentation index, see https://docs.nvidia.com/nemo-helix/llms.txt. > For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.nvidia.com/nemo-helix/_mcp/server. # v0.8.0 - 2026-10-08 NeMo Helix v0.8.0 builds on the 0.7 release line with broader agent optimization and deployment workflows, richer agent-evaluation runners and task discovery, Studio support for optimization and deployment journeys, and platform hardening for setup, SDK clients, deployments, and release operations. ## Highlights * **Expanded agent optimization.** Agents can use new prompt-master and Switchyard optimization strategies, run Pi and remote-agent Fabric harnesses, and carry strategy options through Studio and CLI-backed optimization submissions. * **Broader evaluation execution.** Evaluator adds richer filtering for stored tasks and tasksets, registered-agent execution through Fabric, Gym, and Harbor runners, Files-backed artifact storage for scaled evaluations, and live metric evaluation endpoints. * **Studio optimization and deployment workflows.** Studio can submit optimization jobs, show strategy options and in-progress details, build an agent image before deploying to Docker or Kubernetes, deploy HPO-tuned agent instructions, and guide users when the Helix backend is unreachable or membership is denied. * **Setup, CLI, and SDK hardening.** Setup can create a sample workspace and demo agent, the setup model picker is searchable, CLI startup and HTTP transport reuse are faster, remote-cluster context handling is safer, and product code has moved off the legacy generated `nemo_helix` SDK surface. * **Deployment and builder foundations.** The platform adds container-image builder entities, build/fetch/push/sign steps, deployment HTTP clients, OpenShell-compatible task images and deployment mode support, and model-management paths backed by the deployments API. ## What's included ### Agents and Fabric * Agent optimization gains prompt-master and Switchyard strategy support, plus orchestration between numeric and prompt optimizations. The `nemo agents optimize list-strategies` output includes strategy descriptions, and `nemo agents optimize run-strategy --strategy ` can submit plugin-provided strategies. See [Optimize Agents](/documentation/agents/optimize-agents). * Fabric-backed agents can use the `nvidia.fabric.pi` harness and a remote-agent Fabric adapter. These expand the harnesses available when registering, evaluating, and deploying agents through Fabric-backed workflows. (documentation not yet available) * Agent deployments support the OpenShell deployment mode and expose the set of enabled deployment methods, making deployment choices align with the platform backends available in the current environment. See [Deploy Agents](/documentation/agents/deploy-agents). * Agent execution and analysis jobs can attach extensions that save analysis reports and related artifacts as job results. See [Execute Agents as Jobs](/documentation/agents/execute-agents-as-jobs). ### Agent Evaluation * Stored evaluator tasks and tasksets are now easier to discover: task lists can filter by kind, metric, metadata, tags, and text, and taskset lists can filter by member task and description. See [Filter tasks and tasksets](/documentation/evaluate-models/manage-tasks-tasksets#filter-tasks-and-tasksets). * Agent evaluations can run a registered agent through the Fabric runner, Gym runner, or Harbor runner, so evaluation configs can reference the registered agent instead of embedding a separate config. See [Evaluate Fabric Agents](/documentation/evaluate-models/agent-eval/fabric-runner). * Scaled evaluation artifact storage now routes through the Files service, and Harbor runners install from per-version requirements with OpenSandbox support. This makes evaluation artifacts and runner dependencies align with the platform files and sandboxing model. * Evaluator adds a live metric evaluation endpoint, failed-result representation for intake, better Gym host survivability on cold node image pulls, and shared propagation of Fabric and Gym environment secrets. * `FieldMapping` can match array elements, and evaluator task submission rejects whitespace-only task identifiers earlier. ### Studio * Studio can submit agent optimization jobs from the UI, show optimization strategy options, display additional details while optimization jobs run, and deploy HPO trials that tune an agent's system instructions. * Studio can build an agent image before deploying to Docker or Kubernetes and can deploy newly registered agents using a method the platform can run. * The Run Evaluation flow can select a dataset from a fileset and accepts Parquet and CSV datasets. * Agent analysis flows can send the agent's ETHOS context with analysis runs and expose a **Run analysis now** action. (documentation not yet available) * Studio adds clearer states for unreachable Helix backends, workspace membership authorization failures, and sample workspaces, while preserving recovery actions on unauthorized screens. * Studio improves agent, Customizer, and Data Designer navigation with a trimmed agent details tab, agent registration terminology, a Customizer/Data Designer start-page redesign, saved customization templates, and an Anonymizer entity-mode auto-detect default. ### Customizer and Data Designer * Customizer can serve saved job templates from the customization router, and Studio can start customization jobs from templates. See [Studio customization templates](/documentation/studio/customization#start-from-a-template). * GRPO and DPO training paths are hardened with GRPO offline fixes, rebased NeMo-RL pinning, DTensor checkpoint configuration for DPO, and transformer updates for Unsloth and RL jobs. See [GRPO training](/documentation/customizer-reference/grpo-and-reward-environments). ### Platform, CLI, and Deployment * OIDC configuration now separates public, confidential, and workload client profiles. The CLI supports brokered confidential login without receiving the client secret or provider refresh token. Public CLI clients can opt into the same encrypted server-side session store without becoming confidential, and `platform.advertised_base_url` lets discovery publish the gateway origin when services connect internally. * `nemo setup` can create a sample workspace and demo Fabric agent, and the setup model picker is searchable. See [Setup](/documentation/get-started/setup). * CLI and plugin commands are more remote-cluster aware: skills and config views avoid clobbering configured remote base URLs, denied binds report as permission errors, generated submit commands use the global context, and shared HTTP transports reduce repeated client setup. * Product services, plugins, and tests have moved off the legacy generated `nemo_helix` SDK surface toward typed clients; documentation examples now instantiate clients from the active CLI context. See [Connect with CLI Contexts](/documentation/reference/cli-reference/connect-to-deployments). * Jobs can be resolved by ID as well as by name across get and mutation operations. * Model and deployment integration continues moving through the deployments API, with model create/delete and managed-deployment reads using deployment-backed flows. * The image builder adds image identity records, build requests, backend execution, fetch/build/push/sign steps, image-set filtering, returned image references, prefixed job names, and minikube end-to-end coverage. These are foundational platform capabilities and may not yet have user-facing documentation. ## Maintenance * Release operations preserve version series for pinned nightlies, validate NGC metadata before publishing, and include release-note and release-test-scope skills for local release planning. * Docs now enforce CLI-context connection patterns instead of direct `NHX_BASE_URL` use in examples. * Internal code and tests received reliability improvements for TestClient teardown, free-port usage, sidecar DB access, Alembic percent escaping, and plugin client retry behavior. ## Install For fresh local installs, see [Setup](/documentation/get-started/setup). It covers the recommended PyPI install, source-checkout prerequisites, Flox activation, the system-toolchain fallback, provider configuration, and optional sample workspace creation. For self-managed Kubernetes, start with [Install NeMo Helix Helm Chart](/documentation/kubernetes-deployment/setup/helm/install). ## Upgrade from v0.7.x OIDC client configuration and auth discovery are intentionally breaking in this release. Move flat interactive fields under `auth.oidc.public_client` or `auth.oidc.confidential_client`, move workload fields under `auth.oidc.workload`, and upgrade the server and bundled CLI/SDK together. See [OIDC Setup](/documentation/access-control/authentication/oidc-setup#breaking-configuration-migration) for the complete field mapping. There is no compatibility translation for the old schema. From an existing local checkout: ```bash git fetch git checkout main make bootstrap flox -q activate ``` Without Flox, install the system toolchain versions printed by `make toolchain-versions`, plus a C compiler, then run `make TOOLCHAIN=system bootstrap` followed by `source .venv/bin/activate`. Run setup after updating the checkout: ```bash nemo setup ``` After setup, restart local services before using CLI, SDK, Studio, or plugin workflows against the upgraded checkout. If your automation imported the legacy generated `nemo_helix` SDK package directly, migrate to the typed clients exposed through `nemo_helix_plugin` and the active CLI context before relying on v0.8 workflows. ## Compatibility * Python 3.12, 3.13, or 3.14 * macOS and Linux for local CLI, SDK, Studio, and hosted-provider workflows * Linux x86\_64 for local NVIDIA GPU workloads * CUDA 13-capable NVIDIA drivers for local GPU workloads * Self-managed Kubernetes clusters deployed with Helm * Docker for local services, Docker-backed platform jobs, local ClickHouse, Docker Sandboxes, local model-serving workflows, and Gym task images * Kubernetes/Ray runtime for NeMo-RL DPO and GRPO customization jobs * OpenSandbox and a job-storage PVC for sandboxed Gym and GRPO reward environments * ClickHouse for Intake trace storage and Experiments rollups * NVIDIA GPU access for local training, model serving, RL training, and GPU-backed synthetic data workflows * Node 22.18.0+ for Studio assets * Platform API and `nemo-helix` Python SDK `0.8.0` ## Current constraints * **Self-managed scope.** v0.8.0 documents local setup and user-managed Kubernetes deployment. It is not a managed hosted-service release. * **Research-preview optimizer.** Analyst, Experimentalist, prompt-master, and Switchyard optimization strategies are early agentic workflows. They are useful for guided optimization, but quality and autonomy still require developer review. * **Sandboxed Gym runtime.** Sandboxed Gym evaluation requires Kubernetes or a configured sandbox-capable environment, OpenSandbox setup for cluster runs, and task images that include the environment dependencies. * **GRPO runtime.** GRPO jobs require Kubernetes/Ray execution and sandboxed Gym configuration. They do not have a local Docker fallback. * **OpenShell and image builder scope.** OpenShell deployment support and image-builder capabilities are foundational platform paths and may require environment-specific administrator configuration. ## Links * Repository: [https://github.com/NVIDIA-NeMo/nemo-helix](https://github.com/NVIDIA-NeMo/nemo-helix) * Issues: [https://github.com/NVIDIA-NeMo/nemo-helix/issues](https://github.com/NVIDIA-NeMo/nemo-helix/issues)