feat: support NeMo Relay 0.7

.github/workflows/ci_python.yml

@@ -76,11 +76,11 @@ jobs:
           toolchain: stable
           cache: false
 
-      # Remove nemo-relay once we are using 0.7 (and thus the Python package for nemo-relay-cli)
+      # The Python nemo-relay package does not install the Rust gateway CLI.
       - name: Install test tools
         uses: taiki-e/install-action@c070f87102a1c75b3183910f391c1cb887fe13c8 # v2.77.6
         with:
-          tool: just@1.50.0,nemo-relay-cli@0.6.0
+          tool: just@1.50.0,nemo-relay-cli@0.7.2
 
       - name: Set up uv
         uses: astral-sh/setup-uv@cec208311dfd045dd5311c1add060b2062131d57 # v8

ATTRIBUTIONS-Python.md

@@ -5470,7 +5470,7 @@ Copyright 2016 Andrew Svetlov and aio-libs contributors
    limitations under the License.
 ```
 
-## nemo-relay (0.6.0)
+## nemo-relay (0.7.2)
 
 ### Licenses
 License: `Apache-2.0`

TODO.md

@@ -36,24 +36,6 @@ cleanup validation.
   validate both installed-wheel and standalone Rust CLI behavior before
   removing this entry.
 
-## NeMo Relay 0.6.x Request Decoding Release
-
-- **Status:** Waiting for an upstream release
-- **Added:** July 16, 2026
-- **Affected documentation:** `adapters/codex/README.md` and
-  `docs/integrations/harness/codex.mdx`
-- **Reason:** Released NeMo Relay versions do not yet decode the
-  `zstd`-compressed request bodies emitted by the Codex SDK, so semantic Relay
-  artifacts require a source installation.
-- **Upstream resolution:**
-  [NVIDIA/NeMo-Relay#452](https://github.com/NVIDIA/NeMo-Relay/pull/452), merged
-  as `fe144d0d23e483c8216537118304e306abc20837`
-- **Removal condition:** A published `nemo-relay-cli` version in NeMo Fabric's
-  supported `>=0.6.0,<0.7.0` range contains the merged request-decoding fix.
-- **Cleanup:** Replace the pinned source-install instructions with the released
-  CLI installation, run the Codex Relay end-to-end test, update both affected
-  documentation files, and remove this entry.
-
 # Planned Work
 
 ## Example Authoring Skill

adapters/claude/README.md

@@ -121,8 +121,8 @@ by the SDK and is not persisted as a NeMo Fabric artifact.
 
 ## Relay Observability
 
-Relay requires a separately installed NeMo Relay 0.6.x CLI on `PATH`; the Python
-`nemo-relay` package does not provide the executable. Follow the
+Relay requires a NeMo Relay CLI in the `>=0.7.2,<0.8` range on `PATH`. NeMo
+Fabric's `relay` extra does not install the CLI. Follow the
 [NeMo Relay installation instructions](https://docs.nvidia.com/nemo/fabric/getting-started/install#install-nemo-relay).
 
 Enable Relay through the normalized NeMo Fabric configuration:

adapters/claude/src/nemo_fabric_adapters/claude/adapter.py

@@ -575,12 +575,11 @@ def prepare_claude_relay(payload: dict[str, Any]) -> ClaudeRelaySettings | None:
         ) from error
 
     try:
-        relay_contract = relay_gateway.relay_cli_contract(executable)
+        relay_gateway.relay_cli_contract(executable)
         plugin_config = common_utils.load_relay_plugin_config(payload)
         config_path, plugin_config_path = common_utils.write_relay_configs(
             relay_config={"agents": {"claude": {"command": "claude"}}},
             plugin_config=plugin_config,
-            observability_version=relay_contract.observability_version,
         )
     except (
         OSError,

adapters/codex/README.md

@@ -145,8 +145,8 @@ Codex state variables, the selected model's `api_key_env`, and explicit
 
 ## Relay Integration
 
-Relay requires a separately installed NeMo Relay 0.6.x CLI on `PATH`; the Python
-`nemo-relay` package does not provide the executable. Follow the
+Relay requires a NeMo Relay CLI in the `>=0.7.2,<0.8` range on `PATH`. NeMo
+Fabric's `relay` extra does not install the CLI. Follow the
 [NeMo Relay installation instructions](https://docs.nvidia.com/nemo/fabric/getting-started/install#install-nemo-relay).
 
 Enable Relay with `FabricConfig.enable_relay(...)`. The adapter starts the

adapters/codex/src/nemo_fabric_adapters/codex/adapter.py

@@ -761,7 +761,7 @@ def prepare_codex_relay(
         ) from error
 
     try:
-        relay_contract = relay_gateway.relay_cli_contract(executable)
+        relay_gateway.relay_cli_contract(executable)
         plugin_config = common_utils.load_relay_plugin_config(
             {
                 "agent_name": agent_name,
@@ -774,7 +774,6 @@ def prepare_codex_relay(
             # Codex execution remains SDK-owned; Relay runs only as a gateway.
             relay_config={},
             plugin_config=plugin_config,
-            observability_version=relay_contract.observability_version,
         )
     except (OSError, RuntimeError, ValueError, json.JSONDecodeError) as error:
         raise AdapterRelayError(

adapters/common/src/nemo_fabric_adapters/common/relay_gateway.py

@@ -5,6 +5,7 @@
 
 from __future__ import annotations
 
+import os
 import re
 import shutil
 import socket
@@ -20,8 +21,8 @@
 RELAY_HEALTH_TIMEOUT_SECONDS = 10.0
 RELAY_STOP_TIMEOUT_SECONDS = 5.0
 RELAY_VERSION_TIMEOUT_SECONDS = 5.0
-RELAY_MINIMUM_VERSION = (0, 6, 0)
-RELAY_MAXIMUM_VERSION = (0, 7, 0)
+RELAY_MINIMUM_VERSION = (0, 7, 2)
+RELAY_MAXIMUM_VERSION = (0, 8, 0)
 
 
 class RelayGatewayError(RuntimeError):
@@ -46,7 +47,6 @@ class RelayCliContract:
     """Versioned external Relay CLI contract consumed by Fabric adapters."""
 
     version: tuple[int, int, int]
-    observability_version: int
 
 
 def resolve_relay_command(base_dir: Path, value: str | Path) -> Path:
@@ -87,18 +87,25 @@ def relay_cli_contract(executable: Path) -> RelayCliContract:
         raise RelayGatewayError(
             "NeMo Relay CLI version could not be determined"
         ) from error
-    match = re.search(r"\b(\d+)\.(\d+)\.(\d+)", completed.stdout)
+    match = re.search(
+        r"\b(\d+)\.(\d+)\.(\d+)"
+        r"(-[0-9A-Za-z][0-9A-Za-z.-]*)?"
+        r"(?:\+[0-9A-Za-z][0-9A-Za-z.-]*)?(?=\s|$)",
+        completed.stdout,
+    )
     if completed.returncode != 0 or match is None:
         raise RelayGatewayError("NeMo Relay CLI version could not be determined")
-    major, minor, patch = (int(value) for value in match.groups())
+    major, minor, patch = (int(value) for value in match.groups()[:3])
     version = (major, minor, patch)
-    if not RELAY_MINIMUM_VERSION <= version < RELAY_MAXIMUM_VERSION:
+    if match.group(4) is not None or not (
+        RELAY_MINIMUM_VERSION <= version < RELAY_MAXIMUM_VERSION
+    ):
         raise RelayGatewayError(
             "unsupported NeMo Relay CLI version "
             f"{'.'.join(str(value) for value in version)}; "
-            "NeMo Fabric requires >=0.6.0,<0.7.0"
+            "NeMo Fabric requires >=0.7.2,<0.8.0"
         )
-    return RelayCliContract(version=version, observability_version=2)
+    return RelayCliContract(version=version)
 
 
 def wait_for_relay_gateway(
@@ -169,10 +176,16 @@ def start_relay_gateway(
     if launch.anthropic_base_url is not None:
         command.extend(["--anthropic-base-url", launch.anthropic_base_url])
     try:
+        env = os.environ.copy()
+        # The explicit, runtime-owned plugins.toml is the only non-system
+        # plugin source for this gateway. Keep ambient project config out of
+        # concurrent Fabric runtimes without mutating process-global state.
+        env["NEMO_RELAY_CONFIG_SCOPE"] = "user"
         with launch.log_path.open("wb") as log_stream:
             process = subprocess.Popen(
                 command,
                 cwd=cwd,
+                env=env,
                 stdout=log_stream,
                 stderr=subprocess.STDOUT,
                 start_new_session=True,