NVIDIA / NVIDIA/NemoClaw

[Ubuntu 24.04][Inference] inference set omitting --credential-env on a no-auth compatible-endpoint sandbox fails with contradictory endpoint-url guidance

Open
#10,672 1 comment 0 reactions 1 assignee Claimed by @lyral-commits View on GitHub
area: cli area: inference area: providers area: routing NV QA platform: ubuntu
Dominant language
TypeScript
Stars
22.5k
Forks
3.1k
Avg merge
1d 1h
Merged PRs (30d)
715

Description

## Description

On a sandbox originally onboarded with a no-auth compatible-endpoint provider (no `--credential-env` on record), running `inference set` again while omitting `--credential-env` does not reuse the recorded no-auth binding — it fails claiming a required credential value is missing. The error's own remediation ("omit --endpoint-url") then leads to a second, contradictory error requiring `--endpoint-url`, leaving no working combination of flags for this case.

Platform scope: Reproduced on Ubuntu 24.04 only; other platforms not tested.
Regression: Unknown — earlier versions not tested.

## Environment

```text
OS: Ubuntu 24.04
Architecture: x86_64
Node.js: v22.23.0
npm: 10.9.8
Docker: 29.7.2
OpenShell CLI: 0.0.106
NemoClaw: 0.0.116
OpenClaw: 2026.7.1
```

## Steps to Reproduce

1. Onboard a sandbox against a no-auth compatible-endpoint (no credential-env recorded).
2. Run (omitting `--credential-env`, expecting the recorded no-auth binding to be reused):
```bash
nemoclaw {sandbox} inference set --provider compatible-endpoint --endpoint-url {same-url} --model {model} --inference-api openai-completions
```
3. Observe the error, then retry omitting `--endpoint-url` as its own message suggests.

## Expected Result

Omitting `--credential-env` on an existing compatible-endpoint sandbox reuses whatever credential state (including no-auth) was previously recorded, and the command completes with exit code 0.

## Actual Result

```text
$ nemoclaw {sandbox} inference set --provider compatible-endpoint --endpoint-url {same-url} --model {model} --inference-api openai-completions
endpoint-url is not allowed: HTTPS Pin Runtime adapter requires a non-empty credential value for "{same-url}". This sandbox is already configured for 'compatible-endpoint'. To switch only the model, omit --endpoint-url — inference set reuses the endpoint onboarding already established (the gateway route is not changed by inference set). To point the sandbox at a different endpoint, re-run onboarding with the new endpoint (rebuild reuses the recorded endpoint and cannot change it).
exit code: 2

$ nemoclaw {sandbox} inference set --provider compatible-endpoint --model {model} --inference-api openai-completions
endpoint-url is required for custom-compatible metadata.
exit code: 2
```

## Logs

Not captured beyond the console output shown above.

## Related Bugs

Related but distinct: #10453 (closed) — inference set rejecting a no-auth compatible-endpoint provider as foreign even when parameters exactly match the recorded route. Same general area (no-auth compatible-endpoint credential handling), different specific trigger (explicit matching parameters vs. omitted --credential-env).

Contributor guide

Open the contributing guide

Assessment

This issue has not been assessed yet.

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.