compatible-endpoint.yaml

# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
# SPDX-License-Identifier: Apache-2.0
#
# Forward-looking placeholder. NOT imported by scripts/02-providers.sh today.
#
# The active inference path uses OpenShell's built-in `nvidia` v2 profile via
# `openshell inference set` (see 02-providers.sh) — that's the only way to get
# cluster-inference hardening (streaming timeout, header sanitization, model
# enforcement, gateway-side credential injection) on OpenShell ≤ 0.0.50.
#
# When OpenShell ships "Inference mounting from attached providers"
# (docs/sandboxes/providers-v2.mdx:73 — auto-creates inference.local routes
# from inference_capable v2 profiles), switch 02-providers.sh's inference
# block from `--type nvidia` to `--type nemoclaw-compatible-endpoint` and
# add this file to the per-file `provider profile import` loop. The shape
# below is already correct; only the wiring in 02-providers.sh changes.

id: nemoclaw-compatible-endpoint
display_name: NemoClaw Inference (OpenAI-compatible)
description: OpenAI-compatible inference endpoint; agent calls the upstream host directly with provider-injected API key
category: inference
inference_capable: true
credentials:
  - name: api_key
    description: API key for the OpenAI-compatible inference endpoint
    env_vars: [OPENAI_API_KEY, COMPATIBLE_API_KEY]
    required: true
    auth_style: bearer
    header_name: authorization
endpoints:
  - host: integrate.api.nvidia.com
    port: 443
    protocol: rest
    access: read-write
    enforcement: enforce
  - host: inference-api.nvidia.com
    port: 443
    protocol: rest
    access: read-write
    enforcement: enforce
binaries:
  - /opt/hermes/.venv/bin/python
  - /opt/hermes/.venv/bin/hermes
  - /usr/local/bin/hermes
  - /usr/bin/python3
  - /usr/bin/python3.13
  - /usr/bin/curl