Skip to content

fix: gate local-model onboarding behind NEMOCLAW_EXPERIMENTAL=1#73

Merged
jacobtomlinson merged 2 commits into
mainfrom
fix/hide-local-model-onboarding
Mar 17, 2026
Merged

fix: gate local-model onboarding behind NEMOCLAW_EXPERIMENTAL=1#73
jacobtomlinson merged 2 commits into
mainfrom
fix/hide-local-model-onboarding

Conversation

@ericksoa

@ericksoa ericksoa commented Mar 17, 2026

Copy link
Copy Markdown
Contributor

Summary

Fixes #11 — onboarding presented local-model options (vllm, ollama, NIM) without indicating they're experimental, leading to confusing failures for users expecting a polished experience.

Approach: Show all options in the interactive prompts but clearly label local ones as [experimental]. This encourages experimentation while being transparent. Auto-selection of local providers (the silent misconfiguration path) still requires NEMOCLAW_EXPERIMENTAL=1.

What changed

Component Behavior
Plugin CLI interactive prompt Shows all 5 options; nim-local, vllm, ollama labeled [experimental] in both name and hint
Plugin CLI --endpoint flag Accepts all types; warns "experimental and may not work reliably" for local types, continues
Plugin CLI Ollama auto-detect Only auto-selects (skipping prompt) with NEMOCLAW_EXPERIMENTAL=1
Host CLI inference menu Shows NIM, Ollama, vLLM, install-Ollama with [experimental] labels
Host CLI auto-detection Only auto-selects running vLLM/Ollama with NEMOCLAW_EXPERIMENTAL=1 (no silent misconfiguration)

Example: default interactive prompt

Select your inference endpoint:
  NVIDIA Build (build.nvidia.com)      recommended — zero infra, free credits
  NVIDIA Cloud Partner (NCP)           dedicated capacity, SLA-backed
  Self-hosted NIM [experimental]       experimental — your own NIM container deployment
  Local vLLM [experimental]            experimental — local development
  Local Ollama [experimental]          experimental — localhost:11434

Files changed

  • nemoclaw/src/commands/onboard.ts — labels, warning on --endpoint, gated auto-detect
  • nemoclaw/src/cli.ts — updated --endpoint help text
  • bin/lib/onboard.js[experimental] labels on all local options, gated auto-select

Acceptance criteria from #11

  • Onboarding only presents supported options as primary choices (Build, NCP are unlabeled; others marked experimental)
  • Existing local-model code paths are labeled as experimental
  • The installer does not auto-configure a local provider without user confirmation (auto-select requires NEMOCLAW_EXPERIMENTAL=1)
  • Docs match the actual supported onboarding flow (handled by PR docs: remove docs around local model support #64)

Test results

All 10 test cases passed (static analysis + Docker integration).

# Scenario Assertions Result
1 Plugin CLI prompt shows all 5 options with experimental labels 6 PASS
2 --endpoint vllm warns "experimental" but continues 3 PASS
3 --endpoint build produces no warning 3 PASS
4 Ollama auto-detect blocked without NEMOCLAW_EXPERIMENTAL=1 3 PASS
5 Ollama auto-detect works with NEMOCLAW_EXPERIMENTAL=1 2 PASS
6 Host CLI menu shows all local options with [experimental] labels 7 PASS
7 Host CLI does NOT auto-select without env var 4 PASS
8 Host CLI DOES auto-select with env var 3 PASS
9 CLI help text lists all types, mentions "experimental" 4 PASS
10 Compiled JS contains experimental labels (grep verification) 5 PASS

Total: 40 assertions, 0 failures

ericksoa and others added 2 commits March 16, 2026 21:12
Local inference options (nim-local, vllm, ollama, custom) cause issues
for users but the code paths should remain for advanced users and
development. This gates them behind NEMOCLAW_EXPERIMENTAL=1 so the
default onboarding only presents supported options (build, ncp).

Changes:
- Plugin CLI (onboard.ts): only show build/ncp in interactive prompt;
  reject experimental endpoint types via --endpoint without the env var;
  skip Ollama auto-detection unless experimental
- Host CLI (bin/lib/onboard.js): skip vLLM/Ollama auto-detection and
  hide NIM/Ollama/vLLM/install-Ollama options from the menu unless
  NEMOCLAW_EXPERIMENTAL=1
- CLI help (cli.ts): update --endpoint help text to note experimental

All local code paths still work when NEMOCLAW_EXPERIMENTAL=1 is set.

Closes #11

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Instead of hiding local options entirely, show them in both the
interactive prompt and host CLI menu but label them as [experimental].
This encourages experimentation while being transparent about what's
fully supported.

Changes from previous commit:
- Plugin CLI: always shows all 5 endpoint options, labels nim-local/
  vllm/ollama as [experimental]; --endpoint always accepted with a
  warning for experimental types
- Host CLI: always shows local options in menu with [experimental]
  label; only auto-selects local providers with NEMOCLAW_EXPERIMENTAL=1

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

@jacobtomlinson jacobtomlinson left a comment

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Confirmed end-to-end

@jacobtomlinson jacobtomlinson merged commit 0227e1c into main Mar 17, 2026
jessesanford pushed a commit to jessesanford/NemoClaw that referenced this pull request Mar 24, 2026
…IA#73)

* fix: gate local-model onboarding options behind NEMOCLAW_EXPERIMENTAL=1

Local inference options (nim-local, vllm, ollama, custom) cause issues
for users but the code paths should remain for advanced users and
development. This gates them behind NEMOCLAW_EXPERIMENTAL=1 so the
default onboarding only presents supported options (build, ncp).

Changes:
- Plugin CLI (onboard.ts): only show build/ncp in interactive prompt;
  reject experimental endpoint types via --endpoint without the env var;
  skip Ollama auto-detection unless experimental
- Host CLI (bin/lib/onboard.js): skip vLLM/Ollama auto-detection and
  hide NIM/Ollama/vLLM/install-Ollama options from the menu unless
  NEMOCLAW_EXPERIMENTAL=1
- CLI help (cli.ts): update --endpoint help text to note experimental

All local code paths still work when NEMOCLAW_EXPERIMENTAL=1 is set.

Closes NVIDIA#11

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

* fix: always show experimental options in onboarding, label them clearly

Instead of hiding local options entirely, show them in both the
interactive prompt and host CLI menu but label them as [experimental].
This encourages experimentation while being transparent about what's
fully supported.

Changes from previous commit:
- Plugin CLI: always shows all 5 endpoint options, labels nim-local/
  vllm/ollama as [experimental]; --endpoint always accepted with a
  warning for experimental types
- Host CLI: always shows local options in menu with [experimental]
  label; only auto-selects local providers with NEMOCLAW_EXPERIMENTAL=1

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
mafueee pushed a commit to mafueee/NemoClaw that referenced this pull request Mar 28, 2026
@wscurran wscurran added the bug-fix PR fixes a bug or regression label Jun 8, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

bug-fix PR fixes a bug or regression

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Remove unsupported local-model choices from onboarding until local inference is actually supported

3 participants