update super endpoints to build.nvidia

[AjayThorve]

No description provided.

[greptile-apps]

Greptile Summary

This PR migrates all Nemotron Super 120B references from a temporary NVCF invocation endpoint (5582751c-96c0-47e6-b076-af816ee781dd.invocation.api.nvcf.nvidia.com) and a temp- prefixed model name to the official integrate.api.nvidia.com endpoint with the proper model name nvidia/nemotron-3-super-120b-a12b. Additionally, because the Build API currently has limited availability for this model, the Nemotron Super LLM block is commented out across all configs and notebooks, and nemotron_nano_llm is set as the new default for the researcher, clarifier, and planner roles. A secondary behavioral change enables parallel_tool_calls=True in the clarifier agent.

Key changes:

• All 4 main workflow configs, 3 benchmark configs, 3 notebooks, and numerous documentation pages updated to use the correct integrate.api.nvidia.com endpoint for Nemotron Super (previously a temporary per-instance NVCF URL)
• nemotron_super_llm is commented out in every config file with an explanatory comment; nemotron_nano_llm becomes the default for clarifier_agent.llm, clarifier_agent.planner_llm, and deep_research_agent.researcher_llm
• README.md removes the pre-release "Active Development Branch" warning and introduces a minor duplicate model entry in the Software Components list (nano-30b listed twice)
• src/aiq_agent/agents/clarifier/agent.py enables parallel_tool_calls=True, allowing the clarifier to invoke multiple context-gathering tools concurrently

Confidence Score: 4/5

• Safe to merge — changes are primarily endpoint URL updates and config defaults; the one behavioral code change is low-risk.
• The vast majority of changes are mechanical config/doc updates (endpoint URL swap, model name correction, commenting out a block). The only code change — enabling parallel tool calls in the clarifier — is contained within a well-structured LangGraph node and the existing ToolMessage handling logic is compatible with parallel results. The minor README duplicate entry is cosmetic and does not affect functionality.
• README.md (duplicate model entry at lines 76–78); src/aiq_agent/agents/clarifier/agent.py (parallel_tool_calls behavioral change)

Important Files Changed

Filename	Overview
configs/config_cli_default.yml	Replaced temporary NVCF endpoint for nemotron_super with the public integrate.api.nvidia.com endpoint and commented out the entire block; all agent references updated to use nemotron_nano_llm as the default, with inline comments pointing to the optional Super.
configs/config_frontier_models.yml	Same Super-LLM endpoint migration and comment-out as other configs; deep_research_agent.researcher_llm now defaults to nemotron_nano_llm.
configs/config_web_default_llamaindex.yml	Consistent update: Super LLM commented out, Nano set as default for clarifier and deep research roles.
configs/config_web_frag.yml	Consistent update: Super LLM commented out, Nano set as default for clarifier and deep research roles.
src/aiq_agent/agents/clarifier/agent.py	Changed parallel_tool_calls=False to True in _build_graph; enables concurrent tool invocations in the clarifier workflow — a behavioral improvement but worth confirming with integration tests.
README.md	Removed the develop-branch warning, updated model table and config descriptions to reflect Nano as default and Super as optional — but introduces a duplicate nemotron-3-nano-30b-a3b entry in the Software Components list.
frontends/benchmarks/deepresearch_bench/configs/config_deep_research_bench.yml	Migrated from NVCF temp endpoint to integrate.api.nvidia.com for all models; Super LLM commented out, researcher defaults to Nano.
frontends/benchmarks/deepsearch_qa/configs/config_deepsearch_qa.yml	Same endpoint migration and Super LLM comment-out as the other benchmark configs.
frontends/benchmarks/freshqa/configs/config_full_workflow.yml	Same endpoint migration and Super LLM comment-out; intent classifier and shallow/deep research agents now all use Nano by default.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[User Query] --> B[clarifier_agent\nnow uses nemotron_nano_llm]
    B --> C{Tool calls\nparallel_tool_calls=True}
    C -- "parallel tools" --> D[ToolNode\nmultiple ToolMessages]
    D --> B
    C -- "clarification needed" --> E[ask_for_clarification node]
    E --> B
    C -- "complete" --> F{enable_plan_approval?}
    F -- yes --> G[plan_preview node]
    F -- no --> H[END]
    G --> H

    B2[deep_research_agent\nresearcher_llm] --> I{API Endpoint}
    I -- "default\nnemotron_nano_llm" --> J["integrate.api.nvidia.com\nnvidia/nemotron-3-nano-30b-a3b"]
    I -- "optional\nnemotron_super_llm\ncommented out" --> K["integrate.api.nvidia.com\nnvidia/nemotron-3-super-120b-a12b\n⚠️ limited availability"]

Loading

Comments Outside Diff (2)

1. README.md, line 76-78 (link)

   Duplicate model entry

   nemotron-3-nano-30b-a3b appears twice in the software components list — once as (agents, researcher) and again as (intent classifier). These should be merged into a single entry to avoid confusion.

2. src/aiq_agent/agents/clarifier/agent.py, line 521 (link)

   Parallel tool calls enabled

   This changes parallel_tool_calls from False to True, which allows the LLM to invoke multiple tools simultaneously within the clarifier agent. This is a behavioral change worth confirming: with parallel tool calls, the ToolNode will return multiple ToolMessage objects, and the agent_node check isinstance(state.messages[-1], ToolMessage) correctly handles this since the last element will still be a ToolMessage. Ensure this has been tested with the clarifier's multi-turn flow, as concurrent web searches could occasionally produce conflicting or redundant context that the LLM must reconcile before generating the JSON clarification response.

_{Last reviewed commit: 58ab19b}

[raykallen]

lgtm. we'll patch update for super when ready.