Forward-merge release/2.2 into develop

deploy/helm/README.md

@@ -363,6 +363,15 @@ helm upgrade --install aiq aiq2-web-2.0.0.tgz -n ns-aiq \
 | `SERPER_API_KEY` | Serper API key for Google search |
 | `JINA_API_KEY` | Jina API key |
 | `WANDB_API_KEY` | Weights & Biases API key |
+| `MODAL_TOKEN_ID` | Modal sandbox token ID |
+| `MODAL_TOKEN_SECRET` | Modal sandbox token secret |
+| `AIQ_ARTIFACT_BLOB_PROVIDER` | Artifact byte storage provider; unset or `sql` keeps bytes in SQL, `s3` uses S3-compatible storage |
+| `AIQ_ARTIFACT_S3_BUCKET` | Required when `AIQ_ARTIFACT_BLOB_PROVIDER=s3` |
+| `AIQ_ARTIFACT_S3_ENDPOINT_URL` | Unset for AWS S3; set for MinIO or compatible storage |
+| `AIQ_ARTIFACT_S3_REGION` | Optional S3 region |
+| `AIQ_ARTIFACT_S3_PREFIX` | Optional object-key prefix; defaults to `artifacts/v1` |
+| `AWS_ACCESS_KEY_ID` | Optional S3 access key |
+| `AWS_SECRET_ACCESS_KEY` | Optional S3 secret key |
 
 ### Updating secrets
 

docs/source/examples/cli-with-local-nims.md

@@ -202,7 +202,8 @@ The CLI script starts an interactive session. Type your research query and the s
 
 ## Tips for Local NIMs
 
-- **GPU memory**: Monitor with `nvidia-smi`. The 30B model needs ~40 GB VRAM.
+- **GPU memory**: Monitor with `nvidia-smi`. Size GPUs for the
+  `nvidia/nemotron-3-super-120b-a12b` NIM using the model card and support matrix.
 - **Startup time**: NIM containers take 2--5 minutes to load the model on first start. Wait until `/v1/models` returns a response.
 - **Multiple GPUs**: Use `--gpus '"device=0,1"'` to spread across GPUs, or run separate containers per GPU for different model roles.
 - **Networking**: If running inside Docker Compose, use container names instead of `localhost` for `base_url`.

mcp/pyproject.toml

@@ -71,7 +71,7 @@ override-dependencies = [
     "mako>=1.3.12,<2",
     "nltk>=3.10.0,<4",
     "pillow>=12.2.0,<13",
-    "pyasn1>=0.6.3,<0.7",
+    "pyasn1>=0.6.4,<0.7",
     "pygments>=2.20.0,<3",
     "pyopenssl>=26.0.0,<27",
     "tornado>=6.5.5",

skills/aiq-deploy/SKILL.md

@@ -24,7 +24,7 @@ allowed-tools: Read Bash
 
 # AIQ Deploy Skill
 
-## Purpose
+## When to Use This Skill
 
 Use this skill to get a local or self-hosted NVIDIA AI-Q Blueprint server running and verified for use by
 `aiq-research`.
@@ -59,7 +59,7 @@ git check-ignore deploy/.env
 Expected output: `deploy/.env` or a matching ignore rule. If it is not ignored, stop and fix the ignore rule before
 placing credentials in the file.
 
-## Instructions
+## Workflow
 
 1. Locate or clone the AI-Q repository.
 2. Confirm the expected repository files exist.

skills/aiq-deploy/evals/evals.json

@@ -27,5 +27,17 @@
       "Runs basic validation",
       "Reports the verified AIQ_SERVER_URL"
     ]
+  },
+  {
+    "id": "aiq-deploy-003-negative-research-request",
+    "question": "Use AI-Q to write a deep research report about GPU memory optimization.",
+    "expected_skill": null,
+    "expected_script": null,
+    "ground_truth": "The agent does not use aiq-deploy for a research request. It should use aiq-research if a backend is already available, or only route to deployment if no backend exists.",
+    "expected_behavior": [
+      "Does not route to aiq-deploy as the primary skill",
+      "Does not ask for deployment mode immediately",
+      "Does not start services before checking whether an AI-Q backend is available"
+    ]
   }
 ]

skills/aiq-research/SKILL.md

@@ -32,7 +32,7 @@ allowed-tools: Read Bash
 
 # AIQ Research Skill
 
-## Purpose
+## When to Use This Skill
 
 Use this skill to call a locally running NVIDIA AI-Q Blueprint server through the helper script at
 `scripts/aiq.py`.
@@ -64,7 +64,7 @@ Users need:
 
 The helper script has no third-party Python package dependencies; it uses Python standard-library HTTP modules.
 
-## Instructions
+## Workflow
 
 1. Resolve the target backend URL.
 2. Run `health` before sending research requests.

skills/aiq-research/evals/evals.json

@@ -42,5 +42,17 @@
       "Returns a concise answer for Santa Clara, CA",
       "Does not force a deep_researcher job unless AI-Q returns an async job ID"
     ]
+  },
+  {
+    "id": "aiq-research-004-negative-deployment-request",
+    "question": "Set up AI-Q locally with Docker Compose and verify the backend health endpoint.",
+    "expected_skill": null,
+    "expected_script": null,
+    "ground_truth": "The agent does not use aiq-research for deployment or health setup. It should use aiq-deploy.",
+    "expected_behavior": [
+      "Does not route to aiq-research",
+      "Does not call scripts/aiq.py for the initial setup",
+      "Recognizes this as a deployment and validation request"
+    ]
   }
 ]