NVIDIA/NeMo-Agent-Toolkit v1.9.0

v1.9.0

Repository: NVIDIA/NeMo-Agent-Toolkit

Tag: v1.9.0

Published: 2026-09-10T22:12:09Z

Prerelease: no

Release notes:

What's Changed

🚨 Breaking Changes

• chore: Remove local_sandbox by @dagardner-nv in https://github.com/NVIDIA/NeMo-Agent-Toolkit/pull/2194
• fix: Don't expose the user_id parameter to the LLM by @dagardner-nv in https://github.com/NVIDIA/NeMo-Agent-Toolkit/pull/2190
• Improved user identity resolution by @dagardner-nv in https://github.com/NVIDIA/NeMo-Agent-Toolkit/pull/2197

✨ New Features

• feat(middleware): add HITLMiddleware for human-in-the-loop function interception by @ericevans-nv in https://github.com/NVIDIA/NeMo-Agent-Toolkit/pull/2060
• Enable preflight authentication for applicable authentication providers by @ericevans-nv in https://github.com/NVIDIA/NeMo-Agent-Toolkit/pull/2078
• Track LangChain Runnable callbacks by @WilliamK112 in https://github.com/NVIDIA/NeMo-Agent-Toolkit/pull/2100
• Add MLflow OTLP telemetry exporter, docs, and example by @EnesYilmazcode in https://github.com/NVIDIA/NeMo-Agent-Toolkit/pull/2112
• feat(plugin-api): export runtime context and interactive HITL models by @DABH in https://github.com/NVIDIA/NeMo-Agent-Toolkit/pull/2113
• feat(core): add opt-in provider hooks for generated ids and timestamps by @DABH in https://github.com/NVIDIA/NeMo-Agent-Toolkit/pull/2114
• feat(plugin-api): export the interactive prompt content models by @DABH in https://github.com/NVIDIA/NeMo-Agent-Toolkit/pull/2145
• feat(core): route interaction prompt ids and timestamps via providers by @DABH in https://github.com/NVIDIA/NeMo-Agent-Toolkit/pull/2144
• feat(core): Add CircuitBreakerMiddleware for tool fault tolerance by @sankhyanreyansh in https://github.com/NVIDIA/NeMo-Agent-Toolkit/pull/2158

🔧 Improvements

• Update the wheel upload script by @dagardner-nv in https://github.com/NVIDIA/NeMo-Agent-Toolkit/pull/2101
• feat(langchain): split provider integrations into optional extras by @marcusds in https://github.com/NVIDIA/NeMo-Agent-Toolkit/pull/1989
• fix(finetune): apply CLI config overrides in finetuning runtime by @pranaysb in https://github.com/NVIDIA/NeMo-Agent-Toolkit/pull/2103
• build(ruff): enforce B019 to prevent lru_cache/cache memory leaks by @yczhang-nv in https://github.com/NVIDIA/NeMo-Agent-Toolkit/pull/2106
• Add option to enable redirect-based OAuth flow by @thepatrickchin in https://github.com/NVIDIA/NeMo-Agent-Toolkit/pull/1835
• chore: bump nat-ui submodule to 0297fb14 by @thepatrickchin in https://github.com/NVIDIA/NeMo-Agent-Toolkit/pull/2120
• fix(cache-middleware): bound cache size and enforce safe fuzzy threshold by @ColinM-sys in https://github.com/NVIDIA/NeMo-Agent-Toolkit/pull/1879
• Forward-merge release/1.8 into develop by @willkill07 in https://github.com/NVIDIA/NeMo-Agent-Toolkit/pull/2130
• ci(pre-commit): scope pre-commit checks to PR changes by @willkill07 in https://github.com/NVIDIA/NeMo-Agent-Toolkit/pull/2139
• build(core): require cryptography 48 by @DABH in https://github.com/NVIDIA/NeMo-Agent-Toolkit/pull/2149
• feat(core): make JobStore connection pooling configurable by @jonthedecepticon in https://github.com/NVIDIA/NeMo-Agent-Toolkit/pull/2150
• fix: improve content safety guard result handling by @ericevans-nv in https://github.com/NVIDIA/NeMo-Agent-Toolkit/pull/2180
• feat: Add a new boolean config fail_closed to output_verifier_tools by @dagardner-nv in https://github.com/NVIDIA/NeMo-Agent-Toolkit/pull/2187
• fix(websocket): bind conversation reconnection to user identity by @ericevans-nv in https://github.com/NVIDIA/NeMo-Agent-Toolkit/pull/2189
• chore: Update models by @dagardner-nv in https://github.com/NVIDIA/NeMo-Agent-Toolkit/pull/2195
• feat(auth): configure WebSocket identity credentials by @ericevans-nv in https://github.com/NVIDIA/NeMo-Agent-Toolkit/pull/2196
• fix(adk): prevent split OpenTelemetry stack by @ericevans-nv in https://github.com/NVIDIA/NeMo-Agent-Toolkit/pull/2209
• Package updates by @dagardner-nv in https://github.com/NVIDIA/NeMo-Agent-Toolkit/pull/2216

🐛 Bug Fixes

• fix(api-server): Fix OpenAI chat completions tool-call history validation by @yczhang-nv in https://github.com/NVIDIA/NeMo-Agent-Toolkit/pull/2077
• fix(agent): flatten list-style message content for Bedrock/Anthropic by @ctrlsam in https://github.com/NVIDIA/NeMo-Agent-Toolkit/pull/2064
• fix(latency_sensitivity_demo): wire draft_response fan-in as a barrier by @bbednarski9 in https://github.com/NVIDIA/NeMo-Agent-Toolkit/pull/2086
• fix(multi_frameworks): correct supervisor routing by @bbednarski9 in https://github.com/NVIDIA/NeMo-Agent-Toolkit/pull/2089
• fix(router_agent): terminate graph after branch execution by @bbednarski9 in https://github.com/NVIDIA/NeMo-Agent-Toolkit/pull/2088
• fix(auto_desc): reduce parallel batch summaries by @bbednarski9 in https://github.com/NVIDIA/NeMo-Agent-Toolkit/pull/2090
• fix(mcp): tolerate workflows without aliases by @WilliamK112 in https://github.com/NVIDIA/NeMo-Agent-Toolkit/pull/2097
• Honor ChatRequest.model in LangChain react_agent workflows by @marcusds in https://github.com/NVIDIA/NeMo-Agent-Toolkit/pull/2036
• fix memory leak lru cache by @pranaysb in https://github.com/NVIDIA/NeMo-Agent-Toolkit/pull/2105
• fix(mcp): sanitize parameter names from upstream MCP tool schemas by @zamal-db in https://github.com/NVIDIA/NeMo-Agent-Toolkit/pull/2079
• fix: support function workflows in FastMCP front end by @kiwigitops in https://github.com/NVIDIA/NeMo-Agent-Toolkit/pull/1993
• Fix MessageValidator default message metadata by @fallintoplace in https://github.com/NVIDIA/NeMo-Agent-Toolkit/pull/2011
• fix(core): resolve lru_cache memory retention in DiscoveryMetadata by @pranaysb in https://github.com/NVIDIA/NeMo-Agent-Toolkit/pull/2118
• fix(core): resolve string annotations against the defining module in FunctionInfo by @DABH in https://github.com/NVIDIA/NeMo-Agent-Toolkit/pull/2115
• fix(mcp): recover the MCP client when the lifecycle worker dies by @Hokonoken in https://github.com/NVIDIA/NeMo-Agent-Toolkit/pull/2121
• fix(profiler): import the CrewAI profiler handler from its real module by @EnesYilmazcode in https://github.com/NVIDIA/NeMo-Agent-Toolkit/pull/2128
• fix(profiler): deduplicate Semantic Kernel patch targets by @jstar0 in https://github.com/NVIDIA/NeMo-Agent-Toolkit/pull/2131
• fix(langchain): pass conversation_id as LangGraph thread_id by @Rahulreddy1020 in https://github.com/NVIDIA/NeMo-Agent-Toolkit/pull/2074

*...