0.2.0

What's Changed

✨ New Features

• feat: add customizer as a plugin - AALGO-215 by @soluwalana in #13
• feat(studio): Add Column toggling to GroupExperimentDataView by @nakolean in #269
• feat(studio): v0 experiment detail page by @walston in #260
• feat(studio): Skill management for Code Agent by @htolentino-nvidia in #261
• feat(experiments): Require experiments to be associated with an experiment group by @shanaiabuggy in #268
• feat(plat): Add Python 3.14 support by @matthewgrossman in #237
• feat: endpoint to group spans by trace_id and session_id by @BrianNewsom in #272
• feat(studio): client-side dataset quality checks on JSONL upload (ASTD-73) by @aray12 in #244
• feat(studio): Support CSV/Parquet in dataset schema detection by @steramae-nvidia in #238
• feat(experiments): Roll up agent_names/agent_versions; Span attribute cleanup by @shanaiabuggy in #294
• feat(studio): Add input and tools for multiple chat interactions with Code Agent by @htolentino-nvidia in #257
• feat(studio): add column visibility and filters to test cases table by @walston in #316
• feat(studio): Add CLI/agent instructions to experiment group empty state by @walston in #279
• feat(studio): bulk delete agents from the table by @marcusds in #320
• feat(experiments): Experiments UI misc QA by @shanaiabuggy in #296
• feat(models): Add prompt entity by @steramae-nvidia in #249
• feat: prepare helm chart for kubernetes deployment options by @crookedstorm in #319
• feat(studio): Model Compare route, metrics, input toggle, dataset picker by @nv-odrulea in #204
• feat(studio): Add search/filter to experiment group list by @nakolean in #329
• feat(studio): choose model when creating an example agent by @marcusds in #317
• feat(studio): clone an agent with a new name and model (ASTD-240) by @marcusds in #343
• feat(studio): Improve prompting on the Studio Code Agent by @htolentino-nvidia in #347
• feat(studio): Coding agents assets panel by @dmariali in #333
• feat(unsloth): support Nemotron-H hybrid MoE fine-tuning on DGX Spark by @albcui in #338
• feat(evaluator): add agent-eval SDK domain model value types by @arpitsardhana in #339
• feat(evaluator): add agent-eval engine (AgentEvaluator) by @arpitsardhana in #369
• feat(jobs): auth e2e testing and http_client DI by @ironcommit in #377
• feat(studio): Support images in chat by @steramae-nvidia in #364
• feat(experiments): Add BE for new Experiment pinned_at field by @shanaiabuggy in #349
• feat(studio): link experiment test cases to trace detail view by @walston in #353
• feat(evaluator): persist metrics as entities with CRUD, SDK, and job refs by @SandyChapman in #379
• feat(evaluator): add Runtimes and ProfBench agent-eval examples by @arpitsardhana in #380
• feat(sdk): Add a generic NemoClient by @matthewgrossman in #370
• feat(deployments): scaffold plugin API and registry (AIRCORE-755) by @tylersbray in #280
• feat(models): use nmp-api image as the model-weights puller by @benmccown in #358
• feat(evaluator): promote generic agentic runtimes into the Agent-Eval SDK by @arpitsardhana in #383
• feat(sdk): ParamSpec-based typed HTTP endpoints with method() convenience layer by @matthewgrossman in #397
• ci(release): declare publishable container images for release by @a2bondar in #318
• feat(studio): Coding agent split selection by @dmariali in #435
• feat(jobs): add controller diagnostics and per-module E2E config pooling by @ironcommit in #440
• feat(models): vllm k8s support by @benmccown in #305
• feat(experiments): sort the experiments list by rollup metrics by @shanaiabuggy in #448
• feat(evaluator): inline + hybrid metric bundlers; built-ins skip cloudpickle by @SandyChapman in #438
• feat(evaluator): add Trial→Intake boundary mapping module (D8) by @SandyChapman in #443
• feat(studio): add chat history naming by @dmariali in #457
• feat(deployments): reconcile controller and prerequisite DAG (AIRCORE-758) by @tylersbray in #315
• feat(client): add pagination, retry, and client-side options to NemoClient by @matthewgrossman in #454
• feat(evaluator): AgentEvalTaskset and filesystem evidence handle by @arpitsardhana in #447
• feat(deployments): move prerequisites to Deployment and harden reconcile lifecycle by @tylersbray in #469
• feat(customizer): expand skill to help evaluate lift on adapters trained by @soluwalana in #362
• feat(studio): Implement designs for intake trace and span views by @rrhyne in #460
• feat(nemo-agents-plugin): add codex adapter to nemo-agents plugin by @mmogallapalli in #472
• feat(evaluator): publish_to_intake entrypoint (D9) by @SandyChapman in #456
• feat(nemo-agents-plugin): add cursor adapter to nemo-agents plugin by @mmogallapalli in #487
• feat(studio): intake trace/span view tweaks by @rrhyne in #489

🔧 Improvements

• refactor(intake): Add trace_index MV to efficiently query trace and s… by @BrianNewsom in #282
• refactor(studio): move getTextWithCount to common by @aray12 in #267
• refactor(studio): Remove old files by @steramae-nvidia in #330
• refactor(studio): decompose 11 large components into focused modules (ASTD-229) by @marcusds in #381
• refactor(studio): split routes/index into per-domain route groups by @marcusds in #355

🐛 Bug Fixes

• fix(auditor): garak api upgrade to v0.15.1 by @parkanzky in #241
• fix(studio): SDK fix for guardrails by @steramae-nvidia in #273
• fix(e2e): Respect nemo envvars during e2e tests by @matthewgrossman in #276
• fix: add safe-synthesizer auth/authz by @mckornfield in #224
• fix: Data Designer UX improvements by @mikeknep in #278
• fix(anonymizer): Update tutorial by @anastasia-nesterenko in #240
• fix(lint): Include nmp-common check in standard lint, move authz to nmp-common by @matthewgrossman in #284
• fix(studio): Guardrails types by @steramae-nvidia in #285
• fix(studio): match empty state for fileset table to header by @aray12 in #266
• fix: disable auth by default for local env by @marcusds in #289
• fix: default agents list commands to table output by @drazvan in #297
• fix: add agents service authz by @mckornfield in #300
• fix: Add missing PDP authorization contributions for plugins by @mckornfield in #301
• fix(studio): Experiment Session, expand scores into unique columns by @nakolean in #299
• fix(studio): ExperimentGroupsDataView expand score columns by @nakolean in #298
• fix(customizer): update skill to explain how to operate with and without auth by @soluwalana in #304
• fix: correct a context issue with the automodel build by @crookedstorm in #309
• fix(studio): Code agent skill cards small fixes by @htolentino-nvidia in #310
• fix(studio): only onboard on the first example agent by @marcusds in #307
• fix(jobs): resolve default task image from platform config on K8s by @matthewgrossman in #302
• fix(guardrails): update jailbreak-detect model deployment to match ne… by @albcui in #287
• fix(docs): align Fern url basepath with custom-domain for multi-source by @lbliii in #275
• fix(studio): EvaluationResultsDataView by @steramae-nvidia in #283
• fix(studio): Consolidate StatusBadge logic by @steramae-nvidia in #326
• fix: bearer token validate when auth enabled, resolve nat_runner by @mckornfield in #328
• fix(models): hardcoded adapters sidecar img by @benmccown in #325
• fix(studio): refresh agents table immediately after creating an agent by @marcusds in #323
• fix(tests): Mock HF in integration tests by @matthewgrossman in #259
• fix: adjust imagepullsecret handling by @crookedstorm in #344
• fix(k8s): Move scripts and values over into nemo-platform by @matthewgrossman in #342
• fix(intake): Do not read unneeded trace input/output for list view by @BrianNewsom in #334
• fix(studio): stabilize query cache key for telemetry fetch by @aray12 in #340
• fix(studio): Align on a single test suffix pattern by @steramae-nvidia in #327
• fix: set permissions for nat_runner folders by @mckornfield in #350
• fix: docs platform-all install instructions, remove compose/helm ref by @mckornfield in #351
• fix(customizer): set correct flag for full-sft using Unsloth backend by @anubhutivyas in #346
• fix: correct fallback responses for AUT by @mckornfield in #352
• fix: helm chart KEK secret management must accept ready-made secret by @crookedstorm in #354
• fix(studio): Delete StudioDataGrid by @steramae-nvidia in #357
• fix: remove evaluate this model button by @mckornfield in #363
• fix: correct customizer values in the helm chart by @crookedstorm in #365
• fix(cli): Fix startup banner display by @anastasia-nesterenko in #374
• fix(studio): Test models context menu option is not behind a FF by @htolentino-nvidia in #372
• fix(studio): Add base models to Studio docs by @htolentino-nvidia in #373
• fix: show why modal disabled in studio by @mckornfield in #378
• fix(models): eagerly register LoRA adapters with vLLM by @albcui in #382
• fix(studio): Consolidate Studio console by @steramae-nvidia in #308
• fix: remove tokens from local file on logout by @mckornfield in #388
• fix: keep RC bundles off GitHub Releases by @svvarom in #386
• fix(auth): Fix expired JWT check by @anastasia-nesterenko in #389
• fix(studio): replace missing MetricBundleInput type by @nv-odrulea in #392
• fix(e2e): fix local minikube scripts and switch CLI health checks to /status by @matthewgrossman in #393
• fix(studio): Remove title for group experiments dataview by @nakolean in #404
• fix(intake): Fix POST /evaluator-results idempotentcy by @shanaiabuggy in #410
• fix(intake): Derive evaluator_results name from Harbor reward keys by @shanaiabuggy in #416
• fix: ASTD-254: only upload files for local storage filesets by @marcusds in #418
• fix(studio): Reenable or delete skipped tests by @steramae-nvidia in #400
• fix(docs): update IGW target to be consistent with other docs by @parkanzky in #421
• fix: upgrade all opentelemetry packages to 0.64b0 by @marcusds in #449
• fix(docker): add fastembed-cache-empty placeholder so default bake builds by @SandyChapman in #409
• fix: fix nat crash in nvbug 6336437 by @gabwow in #417
• fix: various docs updates, including hiding intake by @mckornfield in #455
• fix(customizer): CVE remediation by @soluwalana in #452
• fix(customizer): unsloth image CVE resolution by @soluwalana in #464
• fix(studio): Flashing on chat components when Code Agents is thinking by @htolentino-nvidia in #459
• fix: nss image, remove ffmpeg, force wandb version by @mckornfield in #463
• fix(auditor): use async SDK to rewrite IGW when necessary by @parkanzky in #477
• fix(studio): Fix simultaneous user requests not being shown on the Code Agent by @htolentino-nvidia in #473
• fix(studio): Fileset panel animations by @steramae-nvidia in #441
• fix(studio): enforce single-select without callback and apply attributes prop by @marcusds in #486
• fix(studio): tooltip for mean metric columns by @nakolean in #484
• fix(studio): experiment detail view UI polish (ASE-365/366/367) by @walston in #466
• fix(studio): Fix input size on Studio Code Agent and hide artifact panel on landing by @htolentino-nvidia in #485
• fix(a11y): reduce jsx-a11y warnings — common 18→4, studio 45→20 by @marcusds in #446
• fix(customizer): address CVEs in unsloth and automodel by @soluwalana in #474

📝 Documentation

• docs: Cursor Cloud dev environment instructions by @drazvan in #208
• docs: bootstrap first version of customizer plugin docs by @gabwow in #313
• docs(auth): restructure auth docs into Fern navigation and fix rendering issues by @ironcommit in #368
• docs(studio): fix agents.mdx Agent List actions table by @marcusds in #375
• docs(evaluator): fix submit() packager, FilesetRef import, and leaked titles by @SandyChapman in #406
• docs(customizer): update documentation and notebooks for customizer by @soluwalana in #414

🧰 Maintenance

• ci: add advisory Docker build status check for PRs by @svvarom in #200
• ci: pin all versions for third party actions by @crookedstorm in #271
• ci: correct dependencies and logic around dispatch job by @crookedstorm in #274
• chore: Data Designer e2e tests by @mikeknep in #247
• chore(packaging): remove legacy evaluator wiring by @SandyChapman in #230
• ci: scope CodeQL to security queries by @svvarom in #292
• chore: consolidate docker files into a single location and more services by @crookedstorm in #288
• chore: remove ultimately unused docker build elements by @crookedstorm in #293
• chore: improve reliablity of container build by increasing timeout by @crookedstorm in #303
• chore(models): re-enable deployment docs by @benmccown in #306
• ci: replace ytanikin PR labeler with release-drafter autolabeler by @svvarom in #312
• test(intake): derive chat-completions fixture created from now by @svvarom in #321
• chore(customizer): refactor code for unsloth and automodel plugins+services by @anubhutivyas in #242
• chore: upgrade high and med vulns for go and js by @mckornfield in #336
• ci: add docker build in the pipeline to use with a basic k8s install test by @crookedstorm in #345
• chore: add a job to clear the PR cache on merge/close by @crookedstorm in #367
• chore(customizer): automodel plugin - add missing CLI hyperparams by @anubhutivyas in #360
• chore: update readme links to docs page by @mckornfield in #361
• chore: remove more out of date safe-synthesizer docs by @mckornfield in #376
• ci: cache the embed model for the k8s smoke tests by @crookedstorm in #384
• chore: add license URL to csv format, allow output by @mckornfield in #337
• ci: establish workflow to clean up image artifacts by @crookedstorm in #390
• chore: web vulns fixes, remove overrides by @mckornfield in #387
• ci: adjust the behavior of the dependency caching for uv by @crookedstorm in #394
• chore: add documentation for OIDC seed setup by @mckornfield in #402
• ci: use action-semantic-pull-requests directly to update as needed by @crookedstorm in #408
• chore: prepare release notes and version bump by @mckornfield in #403
• chore: more CVE python updates by @mckornfield in #411
• ci: add test builds for the auditor and safe synthesizer containers by @crookedstorm in #423
• ci(release): add manual nightly notification toggle by @svvarom in #442
• chore: update base images in case that helps with anything by @crookedstorm in #445
• chore: update badge for repo to public badge by @mckornfield in #461
• chore: Ensure run_nemo_local uses isolated config by @mikeknep in #458
• test(auth): add integration tests for group-based role binding enforcement by @ironcommit in #468
• chore: bump automodel base to grab wandb upgrade by @mckornfield in #475
• chore(studio): remove 11 unused TSX components by @marcusds in #481

Other Changes

• chore(release): begin 0.1.3 development cycle by @matthewgrossman in #194
• feat(studio): Improve Code assistant's tables by @htolentino-nvidia in #186
• fix(models): extend provider discovery timeout and reduce log noise by @tylersbray in #164
• docs(evaluator): add custom Python metrics tutorial by @SandyChapman in #106
• fix(studio): debounce filter text inputs by @nakolean in #192
• fix: set HF_TOKEN during integration and unit tests by @mckornfield in #202
• docs: Fix NemoGuard casing in Models docs by @JashG in #201
• fix(studio): Docs clarification by @steramae-nvidia in #203
• feat(skills): nemo-model-selection skill with live benchmark cache by @walston in #130
• fix(studio): dataset schema and file browser polish by @nv-odrulea in #181
• feat(guardrails): custom NemoGuard JailbreakDetection model implementation by @albcui in #134
• fix(web): remove pnpm devDependency to unblock Volta users by @yamini in #193
• feat(studio): Experiments list page (FP-182) by @walston in #206
• fix: move nss jobs into plugin, update docs by @mckornfield in #195
• feat(e2e): Add jobs e2e tests w/ fixes, mock inference provider by @matthewgrossman in #179
• feat(intake): Expand filter support on GET /experiments by @shanaiabuggy in #207
• chore(plugin): move image helpers to plugin package by @SandyChapman in #211
• feat(cli): add nemo services rm/prune for ghost instance cleanup by @tylersbray in #172
• feat(auditor): add filtering to target and config list endpoints (AIRCORE-387) by @maxdubrinsky in #182
• feat(studio): Data sample panel by @steramae-nvidia in #187
• docs: Drop Anonymizer quickstart page by @mikeknep in #216
• feat(studio): API to expose test case row level data in experiments by @shanaiabuggy in #198
• feat(agents): typed AgentSpec + explore-first onboarding chain (FP-161) by @aleckhoury in #94
• feat(agents): submit agent evaluate + optimize as jobs (AIRCORE-656) by @marcusds in #66
• feat(studio): Action menu on fileset detail route by @steramae-nvidia in #185
• ci: run nightly releases on weekdays by @svvarom in #214
• fix: generate subtype routes with distinct names by @mckornfield in #219
• fix(studio): Regen SDK schema without duplicate operations by @steramae-nvidia in #220
• test(evaluator): generate cloudpickle example fixtures by @SandyChapman in #212
• fix(packaging): align evaluator SDK extra name by @SandyChapman in #225
• chore(platform-seed): remove evaluator seed dependency by @SandyChapman in #227
• feat(e2e): Add files external provider tests by @matthewgrossman in #217
• fix(studio): improve client-side validation for fileset creation by @aray12 in #213
• feat(filter): support nested-namespace map_entity_field; expose evaluator labels by @maxdubrinsky in #196
• fix(plat): Bind to 127.0.0.1 rather than 0.0.0.0 by @matthewgrossman in #234
• feat(studio): Add ExperimentGroup detail page. by @nakolean in #229
• feat(plugin): Decouple nemo-platform-plugin from nmp-common by @matthewgrossman in #145
• fix(safe-synthesizer): move to 0.1.2 pin fix, update TS generation by @mckornfield in #226
• chore(platform-runner): stop registering legacy evaluator service by @SandyChapman in #228
• feat(studio): experiment group creation modal by @walston in #235
• feat(filter): add InMemoryFilterRepository and use it for jobs status filtering by @maxdubrinsky in #191
• ci: publish RC/stable release bundles as GitHub Release assets by @svvarom in #223
• ci: add an aggregation layer to CI to allow blocking merge more dynamically by @crookedstorm in #239
• ci: move the studio CI tests into ci.yaml so that they can be required by @crookedstorm in #250
• docs: fern docs by @lbliii in #26
• feat(experiments): Create new experiments data seed script by @shanaiabuggy in #248
• fix(sdk): pin InferenceParams to $shared in stainless config by @benmccown in #246
• ci: do not require DCO for repository members by @crookedstorm in #253
• ci: update parallelism of unit tests by @crookedstorm in #254
• docs: update Fern custom domain to docs.nvidia.com/nemo-platform by @lbliii in #255
• fix(studio): wire evaluator plugin SDK + migrate eval UI to EvaluateJob API by @marcusds in #251
• feat(models): vllm deployment in docker by @benmccown in #232
• ci(studio): add SDK gen check for OpenAPI spec changes (ASTD-218) by @marcusds in #245
• chore: make e2e inference test more resilient to race conditions by @mckornfield in #258
• feat(filter): add StringFilter type and widen per-field filter types by @maxdubrinsky in #197
• fix(studio): remove references to nim_deployment after backend change by @dmariali in #262
• ci: add label-driven GitHub release notes for RC/stable releases by @svvarom in #252
• feat(studio): Coding agent studio linking by @dmariali in #236
• feat(studio): Filtering for ExperimentGroup DataView by @nakolean in #243
• ci: pin PR labeler to allowlisted 1.3.0 commit by @svvarom in #265
• fix(auditor): Update auditor usages by @anastasia-nesterenko in #221
• ASTD-260: Gate chat playground on valid deployment target by @marcusds in #415
• ASTD-259: Add missing deps to handleInferFromExisting by @marcusds in #419
• ASTD-257: Guard null firstRow and fix expand button a11y by @marcusds in #420

New Contributors

• @yamini made their first contribution in #193
• @a2bondar made their first contribution in #318
• @rrhyne made their first contribution in #460

Full Changelog: 0.1.2...0.2.0