Set Up an OpenAI-Compatible Endpoint
Set Up an OpenAI-Compatible Endpoint
Use the custom OpenAI-compatible provider for servers that implement /v1/chat/completions or a compatible /v1/responses API.
Examples include vLLM, TensorRT-LLM, llama.cpp, LocalAI, and other compatible servers.
The agent connects to inference.local inside the sandbox.
OpenShell forwards that traffic to the endpoint configured during onboarding.
Start the Server
Start the compatible server before onboarding.
The following example starts vLLM on port 8000.
Port 8000 is the default vLLM host-gateway port.
Set NEMOCLAW_VLLM_PORT before onboarding to select a different port.
For the no-authentication path on bundled host-gateway ports, bind the backend to loopback only. NemoClaw places a token-protected proxy in front of the endpoint. At startup, NemoClaw normally inspects the backend listeners. NemoClaw rejects an observed non-loopback listener because it would bypass the proxy token check. If listener inspection is unavailable, the proxy starts without verifying the backend bind. Onboarding does not display this degraded result. Before you continue, inspect the configured backend port:
Continue only when every listener address is in 127.0.0.0/8, is ::1, or is an IPv4-mapped address in ::ffff:127.0.0.0/8.
Stop the server and correct its bind configuration if the output contains *, 0.0.0.0, ::, or another address.
Endpoints configured with COMPATIBLE_API_KEY use a different authenticated path and do not use this loopback-only proxy requirement.
Run Onboarding
Start the onboard wizard.
Select Other OpenAI-compatible endpoint. Enter the server base URL and the model ID reported by the server.
Use a host-routable URL such as http://localhost:8000/v1 when you want onboarding to verify the API, tool-calling, and streaming paths before gateway registration.
To qualify for automatic rewriting, an HTTP endpoint URL must use the exact loopback host localhost, 127.0.0.1, or [::1].
Automatic rewriting is limited to the port selected by NEMOCLAW_VLLM_PORT (8000 by default) and ports 11434 and 11435.
NemoClaw validates the entered URL from the host and registers the OpenShell gateway route through host.openshell.internal:<port> for sandbox traffic.
Sandbox inference requests continue to use the base inference.local policy, so the managed compatible-endpoint route does not require adding the local-inference preset.
NemoClaw leaves URLs without an explicit port, URLs on :80 or another privileged port, and URLs on unsupported ports unchanged.
Those URLs require a separately compatible runtime topology and network policy.
This rewrite depends on an OpenShell topology that resolves host.openshell.internal inside the sandbox; if that bridge is unavailable, onboarding can still validate the host URL, but nemoclaw <name> status is the authoritative runtime check.
For no-authentication endpoints on the port selected by NEMOCLAW_VLLM_PORT (8000 by default) or port 11434, keep the server bound to loopback.
NemoClaw’s token-protected proxy makes the loopback service reachable from the sandbox without exposing the backend on other host interfaces.
Port 11435 is the default proxy listener, so an endpoint using that port requires COMPATIBLE_API_KEY unless you configure NEMOCLAW_OLLAMA_PROXY_PORT to use a different free port before onboarding.
If you manually enter a sandbox-internal alias such as http://host.openshell.internal:8000/v1, host-side endpoint probing is skipped during onboarding.
Use a host-routable endpoint such as localhost when you need onboarding to verify the API, tool-calling, and streaming paths before gateway registration.
Otherwise, verify the runtime route after onboarding with nemoclaw <name> status and a short agent request.
For an HTTP URL using the exact host localhost, 127.0.0.1, or [::1] and the port selected by NEMOCLAW_VLLM_PORT (8000 by default) or port 11434, the API key prompt says that pressing Enter selects no authentication.
Port 11435 also supports this mode when NEMOCLAW_OLLAMA_PROXY_PORT uses a different free port before onboarding.
Other URLs still require COMPATIBLE_API_KEY.
Refer to Choose a Compatible Inference API for the probe order and runtime API selection.
Supply a Portable Inference Descriptor
A host-side activation component can select a compatible endpoint for the portable experimental profile. The component must write the descriptor before the installer or onboarding command starts. The activation component owns authentication to the descriptor source and writes only the five resolved inference fields.
The descriptor contains an API key.
Use a real /run/nemoclaw directory owned by root or the user who runs NemoClaw.
Do not permit group or other users to write that directory.
Publish the descriptor only at /run/nemoclaw/portable-inference.json.
Use a regular file with one hard link, mode 0600, and ownership by the user who runs NemoClaw.
Do not add descriptor-source locations or source credentials to the descriptor.
Do not include the descriptor or its values in a repository, image, log, shell argument, artifact, or activation-component persistent state.
Use a short-lived API key and set expiresAt to that credential’s expiration time.
Create a temporary regular file in /run/nemoclaw with the required owner and mode 0600.
Write the complete JSON.
Close the temporary file.
Rename that file to /run/nemoclaw/portable-inference.json.
This atomic replacement prevents NemoClaw from reading a partial descriptor.
The descriptor uses this schema:
Each field has one required purpose:
Run the portable installer after the final descriptor is available:
The portable profile creates the openshell-docker network on 10.87.0.0/24 and assigns 169.254.2.2/32 to the host loopback interface.
The address 169.254.2.2 is outside the Portable sandbox subnet and is the host-gateway address for host.openshell.internal.
It carries authenticated mTLS plus sandbox-JWT callbacks to OpenShell on port 8080.
The managed local registry uses the distinct address 10.87.0.3 on port 5000.
Keeping the host-gateway address outside the sandbox subnet prevents a Portable workload from receiving that address.
If Portable onboarding reports that the retired 169.254.1.2/32 address is still assigned to the loopback interface, remove only that exact assignment:
Rerun Portable onboarding:
Do not run the loopback deletion command when onboarding reports another interface or prefix. Investigate and resolve the conflicting assignment before you rerun onboarding.
If Portable onboarding reports that openshell-docker still uses 169.254.1.0/24, use only the commands in that error.
Before it shows a command, NemoClaw verifies the current-user Podman socket, the complete retired network record, and every running or stopped container connected to that network.
Each command includes the verified socket URL and the complete 64-character resource ID.
When no container is connected, onboarding prints one network command:
When the exact NemoClaw-managed registry is the only connected container, onboarding prints the registry commands before the network command. A stopped registry does not need the first command. Run each printed command in order. Continue only after it succeeds:
Do not change the socket or IDs.
Do not add --force to any command.
If inspection is inconclusive or any other container is connected, onboarding prints no removal command.
Leave the network and containers unchanged.
This recovery does not remove sandboxes, inference containers, or Hermes resources.
Use the lifecycle that created each resource.
Then retry onboarding.
Rerun Portable onboarding once the network is gone.
The portable profile handles the descriptor as follows:
OpenShell keeps one active inference route. The local runner is standby only when it already exists. OpenShell does not automatically switch to that runner when the compatible endpoint is unavailable. Use Switch Inference Providers when you need to change the active route.
Endpoint validation is a point-in-time onboarding check, not continuous health monitoring.
During onboarding, NemoClaw sends an authenticated Chat Completions request for the selected model.
After route setup, applicable flows also verify that the sandbox receives non-empty assistant content through inference.local.
These checks do not provide continuous availability monitoring or automatic failover.
NemoClaw consumes and deletes a descriptor only after its file metadata passes these checks.
It deletes an admitted descriptor whether it accepts or rejects the descriptor content.
During onboarding, NemoClaw holds the API key in an asynchronous in-process credential scope instead of process.env.
Unrelated child processes do not inherit the API key.
Compatible-endpoint validation reads the scoped value, and provider registration passes it explicitly to OpenShell.
After registration, OpenShell holds the provider credential and adds it to managed inference requests.
NemoClaw does not write the API key into the sandbox or its persistent state.
The upstream credential expiration still controls the registered credential’s lifetime.
The activation component must supply a new descriptor for each onboarding attempt that needs the compatible endpoint.
Serve a Raw Model File
Start a compatible server for a raw model file instead of passing the file path to NemoClaw.
The Ollama provider accepts Ollama model tags and does not accept a raw .gguf path.
The following example starts llama-server with a GGUF model.
During onboarding, select Other OpenAI-compatible endpoint.
Enter the server base URL and the model ID returned by /v1/models.
Use the model ID, not the raw file path.
For the example above, the server commonly reports NVIDIA-Nemotron3-Nano-4B-Q4_K_M.gguf as its model ID.
Run Non-Interactive Onboarding
Start the endpoint before running the non-interactive command because onboarding validates the server.
Set NEMOCLAW_REASONING=true when the endpoint serves a reasoning-only model.
Reasoning mode validates only /v1/chat/completions and does not verify tool calling or streaming.
Enable it only when the endpoint supports the capabilities your agent requires.
For the raw model example, use the ID returned by /v1/models.
For OpenClaw, NEMOCLAW_REASONING_EFFORT accepts low, medium, high, or default.
A low, medium, or high value writes params.extra_body.reasoning_effort when the selected API is openai-completions.
Another API family omits the field.
An unset value or default leaves the endpoint’s own default in place.
NemoClaw rejects an invalid value or a provider/API mismatch before changing provider, sandbox, policy, or registry state.
After inference set changes the effort, an ordinary sandbox restart preserves the persisted value instead of restoring the image’s original onboarding value.
Private and reserved addresses are blocked by default. To use an inference gateway on a trusted corporate network, list only its exact host and keep the endpoint URL on that host:
NemoClaw still resolves the host before probing and pins outbound validation to the complete canonical address set. An exact trusted host can return both public and supported private addresses. NemoClaw pins every canonical answer. If any answer is a disallowed private, reserved, or special-purpose address, validation rejects the endpoint instead of discarding that answer. Among private answers, NemoClaw admits only RFC1918, carrier-grade network address translation (CGNAT), and IPv6 unique local address (ULA) destinations. Link-local metadata and other reserved ranges remain blocked. An unlisted private host, a hostname suffix match, or a DNS failure also remains blocked.
Related Topics
- Choose a Compatible Inference API to select Chat Completions or Responses.
- Meet Custom Endpoint Security Requirements before saving a public custom endpoint.
- Verify the Inference Route after setup.