diff --git a/environments/prime_general_agent/README.md b/environments/prime_general_agent/README.md
new file mode 100644
index 000000000..3e86bb804
--- /dev/null
+++ b/environments/prime_general_agent/README.md
@@ -0,0 +1,45 @@
+# Description
+
+Self-growing toolbench environment that currently 4,418 synthetic tool-use tasks across 1,037 families and 5 difficulty tiers and can be used to generate more. 
+
+Prime Intellect environment integrated through the verifiers library integration.
+
+Full description: https://app.primeintellect.ai/dashboard/environments/primeintellect/general-agent
+
+For installing additional prime environments from the Environments Hub and generating datasets, see [responses_api_agents/verifiers_agent/README.md](../../responses_api_agents/verifiers_agent/README.md).
+
+## Notes
+
+1. Use the local solver to start, not the default env id. Set `vf_env_id: general-agent-solver-local` in `config.yaml`. The default `general-agent` env id may route to a sandbox solver (including options with OpenCode or RLM) that requires a funded `PRIME_API_KEY`. The local solver runs in-process with no sandbox.
+2. Dataset `info` must be at the row top level. The local solver reads `state["info"]["task_dir"]`. If you have a dataset where `info` is nested under `verifier_metadata`, lift it. `prepare.py` writes rows in the correct shape.
+3. `agent_ref.name` must match the agent directory name (`prime_general_agent`). NeMo Gym dispatches rows to the agent by this name.
+4. Corpus version must match the installed package. `info.task_dir` paths point into the installed `general-agent` package's bundled tasks/ tree. Regenerate the dataset whenever you bump the `general-agent` pin.
+5. The requirements is currently pinned to verifiers main as the regular 0.1.14 pin seems too old, but this should be pinned soon. It is currently tested on verifiers>=0.1.15.dev2. 
+
+## Quick start
+
+```bash
+ng_run "+config_paths=[environments/prime_general_agent/config.yaml,responses_api_models/vllm_model/configs/vllm_model.yaml]"
+```
+
+```bash
+ng_collect_rollouts \
+    +agent_name=prime_general_agent \
+    +input_jsonl_fpath=environments/prime_general_agent/data/example.jsonl \
+    +output_jsonl_fpath=results/prime_general_agent_rollouts.jsonl \
+    +limit=1
+```
+
+## Prepare training data
+
+```bash
+python environments/prime_general_agent/prepare.py --split train --output environments/prime_general_agent/data/train.jsonl
+```
+
+# Licensing information
+Code: Apache 2.0
+Data: N/A
+
+Dependencies
+- nemo_gym: Apache 2.0
+- verifiers: Apache 2.0
diff --git a/environments/prime_general_agent/__init__.py b/environments/prime_general_agent/__init__.py
new file mode 100644
index 000000000..e69de29bb
diff --git a/environments/prime_general_agent/config.yaml b/environments/prime_general_agent/config.yaml
new file mode 100644
index 000000000..1137caa05
--- /dev/null
+++ b/environments/prime_general_agent/config.yaml
@@ -0,0 +1,23 @@
+prime_general_agent:
+  responses_api_agents:
+    prime_general_agent:
+      entrypoint: app.py
+      domain: agent
+      description: Self-growing toolbench environment with synthetic tool-use tasks from Prime Intellect.
+      value: Improve multi-turn tool use across diverse task families.
+      model_server:
+        type: responses_api_models
+        name: policy_model
+      model_name: ""
+      vf_env_id: general-agent-solver-local
+      vf_env_args: {}
+      group_size: 1
+      max_concurrent_generation: -1
+      max_concurrent_scoring: -1
+      max_tokens: 16000
+      temperature: 1.0
+      top_p: 1.0
+      datasets:
+        - name: example
+          type: example
+          jsonl_fpath: environments/prime_general_agent/data/example.jsonl
diff --git a/environments/prime_general_agent/data/.gitignore b/environments/prime_general_agent/data/.gitignore
new file mode 100644
index 000000000..4424b6fde
--- /dev/null
+++ b/environments/prime_general_agent/data/.gitignore
@@ -0,0 +1,5 @@
+*train.jsonl
+*validation.jsonl
+*train_prepare.jsonl
+*validation_prepare.jsonl
+*example_prepare.jsonl
diff --git a/environments/prime_general_agent/data/example.jsonl b/environments/prime_general_agent/data/example.jsonl
new file mode 100644
index 000000000..0aa57e560
--- /dev/null
+++ b/environments/prime_general_agent/data/example.jsonl
@@ -0,0 +1 @@
+{"task_idx": 0, "vf_env_id": "general-agent-solver-local", "agent_ref": {"type": "responses_api_agents", "name": "prime_general_agent"}, "responses_create_params": {"input": []}, "info": {}}
diff --git a/environments/prime_general_agent/prepare.py b/environments/prime_general_agent/prepare.py
new file mode 100644
index 000000000..6babaabff
--- /dev/null
+++ b/environments/prime_general_agent/prepare.py
@@ -0,0 +1,90 @@
+# Copyright (c) 2026, NVIDIA CORPORATION.  All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""Generate a Gym JSONL for the prime general-agent verifiers environment.
+
+The local solver does ``task_dir = Path(state["info"]["task_dir"])`` so the
+``info`` dict must live at the top level of each row (not inside
+``verifier_metadata``). This script writes rows in that lifted shape.
+
+The ``info.task_dir`` paths point into the installed ``general-agent`` package's
+bundled tasks/ tree, so the package version must match the corpus version that
+produced the dataset.
+
+Requires the agent venv (or a venv with verifiers + general-agent installed):
+    pip install 'verifiers @ git+https://github.com/PrimeIntellect-ai/verifiers.git@main' \\
+        --extra-index-url https://hub.primeintellect.ai/primeintellect/simple/
+    pip install general-agent==0.1.4
+
+Usage:
+    python environments/prime_general_agent/prepare.py --output environments/prime_general_agent/data/train.jsonl
+"""
+
+import argparse
+import json
+from pathlib import Path
+
+
+VF_ENV_ID = "general-agent-solver-local"
+AGENT_NAME = "prime_general_agent"
+
+
+def main() -> None:
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--split", default="train", help="verifiers env split (typically 'train' or 'test')")
+    parser.add_argument("--limit", type=int, default=None)
+    parser.add_argument("--output", required=True)
+    args = parser.parse_args()
+
+    try:
+        import verifiers as vf
+    except ImportError:
+        raise ImportError("Install verifiers + general-agent first; see module docstring.")
+
+    env = vf.load_environment(VF_ENV_ID)
+    dataset = env.get_dataset(args.split)
+    rows = list(dataset)
+    if args.limit is not None:
+        rows = rows[: args.limit]
+
+    output_path = Path(args.output)
+    output_path.parent.mkdir(parents=True, exist_ok=True)
+    with output_path.open("w") as f:
+        for idx, row in enumerate(rows):
+            # The local solver reads info.task_dir at the row top level.
+            info = row.get("info")
+            if isinstance(info, str):
+                info = json.loads(info)
+            info = info or {}
+
+            prompt_input = []
+            if "prompt" in row and row["prompt"]:
+                prompt_input.append({"role": "user", "content": row["prompt"]})
+            elif "question" in row and row["question"]:
+                prompt_input.append({"role": "user", "content": row["question"]})
+
+            out = {
+                "task_idx": idx,
+                "vf_env_id": VF_ENV_ID,
+                "agent_ref": {"type": "responses_api_agents", "name": AGENT_NAME},
+                "responses_create_params": {"input": prompt_input},
+                "info": info,
+                **{k: v for k, v in row.items() if k not in {"responses_create_params", "info"}},
+            }
+            f.write(json.dumps(out) + "\n")
+
+    print(f"Wrote {len(rows)} rows to {output_path}")
+
+
+if __name__ == "__main__":
+    main()
diff --git a/responses_api_agents/prime_general_agent/__init__.py b/responses_api_agents/prime_general_agent/__init__.py
new file mode 100644
index 000000000..e69de29bb
diff --git a/responses_api_agents/prime_general_agent/app.py b/responses_api_agents/prime_general_agent/app.py
new file mode 100644
index 000000000..b6da9c79b
--- /dev/null
+++ b/responses_api_agents/prime_general_agent/app.py
@@ -0,0 +1,18 @@
+# Copyright (c) 2026, NVIDIA CORPORATION.  All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+from responses_api_agents.verifiers_agent.app import VerifiersAgent
+
+
+if __name__ == "__main__":
+    VerifiersAgent.run_webserver()
diff --git a/responses_api_agents/prime_general_agent/requirements.txt b/responses_api_agents/prime_general_agent/requirements.txt
new file mode 100644
index 000000000..326eb2471
--- /dev/null
+++ b/responses_api_agents/prime_general_agent/requirements.txt
@@ -0,0 +1,4 @@
+-e nemo-gym[dev] @ ../../
+verifiers @ git+https://github.com/PrimeIntellect-ai/verifiers.git@main
+--extra-index-url https://hub.primeintellect.ai/primeintellect/simple/
+general-agent==0.1.4
diff --git a/responses_api_agents/prime_general_agent/tests/__init__.py b/responses_api_agents/prime_general_agent/tests/__init__.py
new file mode 100644
index 000000000..e69de29bb