fix: offer Ollama install fallback when cloud API is unavailable

[futhgar]

Summary

• Always show an "Install Ollama" option during onboard on Linux (not just macOS), so users have a local inference fallback when build.nvidia.com is overloaded or down
• Use the official curl -fsSL https://ollama.com/install.sh | sh installer on Linux (Homebrew on macOS)
• Add install-ollama as a valid NEMOCLAW_PROVIDER value for non-interactive/CI mode

Problem

When Ollama is not installed and the NVIDIA API server is unavailable, nemoclaw onboard presents only the cloud option on Linux. Users can't get an API key and onboarding fails with no fallback path (issue #301).

Fix

The install-ollama option was already implemented for macOS but gated behind process.platform === "darwin". This PR extends it to Linux with the appropriate installer command. The change is minimal — no restructuring of the onboard flow.

Fixes #301

Test plan

• New test: verifies Install Ollama (Linux) option appears when Ollama is not installed on Linux
• New test: verifies the curl installer (not Homebrew) is invoked on Linux
• Existing onboard-selection.test.js tests still pass
• All onboard/credentials/inference-config tests pass
• npm test shows no new failures (pre-existing failures in runtime-shell.test.js and install-preflight.test.js are unrelated)

Summary by CodeRabbit

Release Notes

• New Features
  • Added automatic Ollama installation option with platform-specific methods: brew install on macOS and curl-based installation on Linux
  • Enhanced provider selection with improved fallback behavior during non-interactive setup

Signed-off-by: Josue Balandrano Coronel josuebc@pm.me

[coderabbitai]

Note

Reviews paused

It looks like this branch is under active development. To avoid overwhelming you with review comments due to an influx of new commits, CodeRabbit has automatically paused this review. You can configure this behavior by changing the reviews.auto_review.auto_pause_after_reviewed_commits setting.

Use the following commands to manage reviews:

• @coderabbitai resume to resume automatic reviews.
• @coderabbitai review to trigger a single review.

Use the checkboxes below for quick actions:

• ▶️ Resume reviews
• 🔍 Trigger review

📝 Walkthrough

Walkthrough

Extended src/lib/onboard.ts to support a new install-ollama provider option with platform-specific installation methods (brew for macOS, curl for Linux), including fallback logic when the exact provider option is unavailable in non-interactive mode.

Changes

Cohort / File(s)	Summary
Install-Ollama Provider Support src/lib/onboard.ts	Added install-ollama provider validation, implemented platform-specific installation logic (brew on macOS, curl on Linux), and introduced graceful fallback to existing ollama option when install-ollama is requested but unavailable in non-interactive mode.

Estimated code review effort

🎯 3 (Moderate) | ⏱️ ~20 minutes

Poem

🐰 A hop, skip, and local install away,
When API servers fail and block the way,
Ollama installs with brew and with curl,
Cross-platform magic, let fallbacks unfurl!
No more brittle paths—graceful they dance,
A second chance for every onboard advance. 🌟

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 75.00% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title directly describes the main change: offering an Ollama install fallback when cloud API is unavailable, which aligns with the primary objective of providing a local inference option when the NVIDIA API server is down.
Linked Issues check	✅ Passed	The PR fully addresses issue #301 by adding the 'install-ollama' provider option for both macOS and Linux with platform-specific installers, enabling local inference when the NVIDIA API is unavailable.
Out of Scope Changes check	✅ Passed	All changes are directly scoped to issue #301: extending non-interactive provider validation, adding install-ollama option generation, updating installation logic, and supporting the fallback behavior for local inference.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[coderabbitai]

Actionable comments posted: 2

🤖 Prompt for all review comments with AI agents

Verify each finding against the current code and only fix it if needed.

Inline comments:
In `@bin/lib/onboard.js`:
- Around line 607-613: The installer calls that invoke run("brew install
ollama", { ignoreError: true }) and run("curl -fsSL
https://ollama.com/install.sh | sh", { ignoreError: true }) must not swallow
failures; update the onboarding logic so the run calls return and/or throw on
error (remove ignoreError: true) and explicitly check the result or catch
exceptions in the surrounding function (the installer branch in onboard.js) and
on failure log a clear error and abort onboarding (process.exit(1) or rethrow)
before proceeding to switch provider/model to Ollama.
- Around line 502-509: The current options builder omits the "install-ollama"
option when Ollama is present (hasOllama/ollamaRunning), causing
CI/non-interactive runs with NEMOCLAW_PROVIDER=install-ollama to fail; modify
the logic that populates options (the block referencing hasOllama, ollamaRunning
and pushing to options) to special-case the environment variable
NEMOCLAW_PROVIDER === "install-ollama" so the "install-ollama" entry is added
unconditionally when that env var is set, or alternatively relax the
non-interactive provider validation to accept "install-ollama" even if the
option wasn't pushed; update references in the same function that validate the
chosen provider to check process.env.NEMOCLAW_PROVIDER as an allowed provider.

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro

Run ID: 69e50485-aeb9-43e5-8f18-c3ea67223c15

📥 Commits

Reviewing files that changed from the base of the PR and between 9513eca and feac390.

📒 Files selected for processing (2)

• bin/lib/onboard.js
• test/onboard-selection.test.js

[coderabbitai]

Actionable comments posted: 1

🧹 Nitpick comments (1)

bin/lib/onboard.js (1)

2216-2260: Minor indentation inconsistency at Line 2226.

The sleep(2) call has 8-space indentation instead of 6, breaking alignment with the surrounding code. This doesn't affect functionality but harms readability.

🧹 Fix indentation

       run("OLLAMA_HOST=0.0.0.0:11434 ollama serve > /dev/null 2>&1 &", { ignoreError: true });
-        sleep(2);
+      sleep(2);

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@bin/lib/onboard.js` around lines 2216 - 2260, In the selected.key ===
"install-ollama" branch adjust the indentation of the sleep(2); line so it
aligns with the surrounding console.log and run calls (same indentation level as
the preceding run("OLLAMA_HOST=...") and console.log lines) — replace the
8-space indent with the 6-space indent used by the block to restore consistent
formatting.

🤖 Prompt for all review comments with AI agents

Verify each finding against the current code and only fix it if needed.

Inline comments:
In `@test/onboard-selection.test.js`:
- Around line 1327-1333: The test is selecting option "2" (OpenAI) instead of
Install Ollama; update the credentials.prompt stub in the test so the first
prompt returns "7" (Install Ollama) instead of "2" to exercise the ollama flow
(reference credentials.prompt and setupNim()), and add a fake curl binary into a
temporary bin and ensure the spawnSync call used by the code under test sees
that bin by updating the spawnSync/env used in the test to include the temp bin
at the front of PATH (reference spawnSync and PATH) so the install-ollama path
can detect curl and proceed.

---

Nitpick comments:
In `@bin/lib/onboard.js`:
- Around line 2216-2260: In the selected.key === "install-ollama" branch adjust
the indentation of the sleep(2); line so it aligns with the surrounding
console.log and run calls (same indentation level as the preceding
run("OLLAMA_HOST=...") and console.log lines) — replace the 8-space indent with
the 6-space indent used by the block to restore consistent formatting.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro

Run ID: 722fed8a-52cc-48e1-a9c7-c02618cb0873

📥 Commits

Reviewing files that changed from the base of the PR and between d94ddea and d40b6fc.

📒 Files selected for processing (2)

• bin/lib/onboard.js
• test/onboard-selection.test.js

[futhgar]

Rebased against main and ported to TypeScript per review feedback. All three original commits were squashed into one since the follow-up fixes (non-interactive fallback, ignoreError removal, test index correction) were incorporated directly into the TypeScript port.

[coderabbitai]

Actionable comments posted: 2

Caution

Some comments are outside the diff and can’t be posted inline due to platform limitations.

⚠️ Outside diff range comments (1)

src/lib/onboard.ts (1)

3306-3366: ⚠️ Potential issue | 🟠 Major

Reuse the existing Ollama startup path here.

This copy of the ollama branch already diverged: it dropped the isWsl() bind workaround, so WSL users who take the new Linux install path will start Ollama on 0.0.0.0 and hit the host-gateway reachability problem the original branch avoids. Extract the shared “start/pick/pull/validate Ollama” flow and keep only the install step platform-specific.

As per coding guidelines, "Limit cyclomatic complexity to 20 in JavaScript/TypeScript files, with target of 15".

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@src/lib/onboard.ts` around lines 3306 - 3366, The install-ollama branch
duplicated the entire Ollama start/pick/pull/validate flow and dropped the
isWsl() bind workaround, increasing complexity and breaking WSL; extract the
shared startup/selection/validation logic into a reusable function (e.g.,
startAndValidateOllama or ensureLocalOllamaReady) and keep the platform-specific
part only for installation steps inside the selected.key === "install-ollama"
branch; ensure the new function uses the existing isWsl() check when building
the run command for starting Ollama (use host vs loopback bind behavior as in
the original flow), reuses symbols like OLLAMA_PORT,
getLocalProviderBaseUrl/getLocalProviderValidationBaseUrl,
getOllamaModelOptions, prepareOllamaModel, validateOpenAiLikeSelection,
promptOllamaModel, requestedModel, BACK_TO_SELECTION, and sets
preferredInferenceApi = "openai-completions" before returning — this both
restores the WSL bind workaround and reduces cyclomatic complexity by moving the
while loop and validation into the shared function.

🧹 Nitpick comments (1)

src/lib/onboard.ts (1)

2880-2890: Add coverage for the silent install-ollama fallback.

This branch only runs in non-interactive mode, and the existing Linux onboarding test in test/onboard-selection.test.ts only covers the “Ollama not installed” case. A small regression test for NEMOCLAW_PROVIDER=install-ollama when Ollama is already present would lock in the new CI behavior.

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@src/lib/onboard.ts` around lines 2880 - 2890, Add a test that exercises the
non-interactive "install-ollama" fallback path when Ollama is already installed:
in test/onboard-selection.test.ts add a case that sets
NEMOCLAW_PROVIDER=install-ollama, mocks/simulates Ollama as present (so the
options list includes a provider with key "ollama"), runs the onboarding
selection flow that uses providerKey and the options.find((o) => o.key ===
"ollama") branch, and asserts the code chooses the existing "ollama" provider
(no process.exit and no error logged). Ensure the test invokes the same
selection entrypoint used by the onboarding code so it hits the branch where
providerKey === "install-ollama" and verifies the silent fallback behavior.

🤖 Prompt for all review comments with AI agents

Verify each finding against the current code and only fix it if needed.

Inline comments:
In `@src/lib/onboard.ts`:
- Around line 1777-1782: The resume conflict arises because setupNim() maps both
"ollama" and "install-ollama" to provider: "ollama-local" but
getEffectiveProviderName() doesn't normalize the "install-ollama" input, causing
getResumeConfigConflicts() to see a mismatch; update getEffectiveProviderName()
to treat the normalized input "install-ollama" the same as "ollama" (i.e.,
return "ollama-local" for either), or add "install-ollama" to the same
mapping/validProviders handling used for normalization so resumed runs with
NEMOCLAW_PROVIDER=install-ollama match the recorded "ollama-local".
- Around line 3311-3312: The piped installer call in onboard.ts using run("curl
-fsSL https://ollama.com/install.sh | sh") can hide curl failures because run()
spawns bash without pipefail; update the call or the run helper so the shell
enables pipefail. Either prepend "set -o pipefail; " (or "set -euo pipefail; ")
to the cmd string passed to run() where Ollama is installed, or modify the run()
implementation in src/lib/runner.ts (the run function that calls
spawnSync("bash", ["-c", cmd])) to always invoke bash with pipefail enabled
(e.g., ensure every "-c" execution sources "set -o pipefail" before the command)
so curl failures cause a non-zero exit.

---

Outside diff comments:
In `@src/lib/onboard.ts`:
- Around line 3306-3366: The install-ollama branch duplicated the entire Ollama
start/pick/pull/validate flow and dropped the isWsl() bind workaround,
increasing complexity and breaking WSL; extract the shared
startup/selection/validation logic into a reusable function (e.g.,
startAndValidateOllama or ensureLocalOllamaReady) and keep the platform-specific
part only for installation steps inside the selected.key === "install-ollama"
branch; ensure the new function uses the existing isWsl() check when building
the run command for starting Ollama (use host vs loopback bind behavior as in
the original flow), reuses symbols like OLLAMA_PORT,
getLocalProviderBaseUrl/getLocalProviderValidationBaseUrl,
getOllamaModelOptions, prepareOllamaModel, validateOpenAiLikeSelection,
promptOllamaModel, requestedModel, BACK_TO_SELECTION, and sets
preferredInferenceApi = "openai-completions" before returning — this both
restores the WSL bind workaround and reduces cyclomatic complexity by moving the
while loop and validation into the shared function.

---

Nitpick comments:
In `@src/lib/onboard.ts`:
- Around line 2880-2890: Add a test that exercises the non-interactive
"install-ollama" fallback path when Ollama is already installed: in
test/onboard-selection.test.ts add a case that sets
NEMOCLAW_PROVIDER=install-ollama, mocks/simulates Ollama as present (so the
options list includes a provider with key "ollama"), runs the onboarding
selection flow that uses providerKey and the options.find((o) => o.key ===
"ollama") branch, and asserts the code chooses the existing "ollama" provider
(no process.exit and no error logged). Ensure the test invokes the same
selection entrypoint used by the onboarding code so it hits the branch where
providerKey === "install-ollama" and verifies the silent fallback behavior.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro Plus

Run ID: 0924340b-f264-4e5d-bef4-d058406d6406

📥 Commits

Reviewing files that changed from the base of the PR and between 8e84410 and 244d62e.

📒 Files selected for processing (1)

• src/lib/onboard.ts

[coderabbitai]

⚠️ Potential issue | 🟠 Major

Map install-ollama to ollama-local for resume checks.

setupNim() records both ollama and install-ollama as provider: "ollama-local", but getEffectiveProviderName() does not normalize the new key. That makes getResumeConfigConflicts() reject a resumed non-interactive run with NEMOCLAW_PROVIDER=install-ollama against a session that already recorded ollama-local.

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@src/lib/onboard.ts` around lines 1777 - 1782, The resume conflict arises
because setupNim() maps both "ollama" and "install-ollama" to provider:
"ollama-local" but getEffectiveProviderName() doesn't normalize the
"install-ollama" input, causing getResumeConfigConflicts() to see a mismatch;
update getEffectiveProviderName() to treat the normalized input "install-ollama"
the same as "ollama" (i.e., return "ollama-local" for either), or add
"install-ollama" to the same mapping/validProviders handling used for
normalization so resumed runs with NEMOCLAW_PROVIDER=install-ollama match the
recorded "ollama-local".

[coderabbitai]

⚠️ Potential issue | 🔴 Critical

🧩 Analysis chain

🏁 Script executed:

#!/bin/bash
set -euo pipefail

RUNNER_FILE="$(fd -i '^runner\.(ts|js)$' src/lib | head -n 1)"
if [[ -z "${RUNNER_FILE}" ]]; then
  echo "Could not locate src/lib/runner.{ts,js}" >&2
  exit 1
fi

echo "Inspecting ${RUNNER_FILE} for run()/runCapture() shell invocation details..."
rg -n -C3 'pipefail|function run\b|const run\b|function runCapture\b|const runCapture\b|spawn(Sync)?\b|exec(File|Sync)?\b|bash|sh -lc|shell:' "$RUNNER_FILE"

echo
echo "Demonstrating default bash pipeline status vs pipefail:"
bash -lc 'false | cat >/dev/null; echo "default pipeline exit: $?"'
bash -o pipefail -lc 'false | cat >/dev/null; echo "pipefail pipeline exit: $?"'

Repository: NVIDIA/NemoClaw

Length of output: 2092

---

Fix the installer command to detect curl failures.

The run() helper in src/lib/runner.ts uses spawnSync("bash", ["-c", cmd]) without set -o pipefail. This means the piped installer command can exit successfully even if curl fails with a download/TLS/HTTP error, because the pipeline will report sh's status instead. The missing curl output leads sh to execute an empty script, exiting 0.

Recommended fixes:

• Add set -o pipefail at the start of the cmd string passed to run()
• Or modify the run() helper to enforce pipefail for all bash invocations

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@src/lib/onboard.ts` around lines 3311 - 3312, The piped installer call in
onboard.ts using run("curl -fsSL https://ollama.com/install.sh | sh") can hide
curl failures because run() spawns bash without pipefail; update the call or the
run helper so the shell enables pipefail. Either prepend "set -o pipefail; " (or
"set -euo pipefail; ") to the cmd string passed to run() where Ollama is
installed, or modify the run() implementation in src/lib/runner.ts (the run
function that calls spawnSync("bash", ["-c", cmd])) to always invoke bash with
pipefail enabled (e.g., ensure every "-c" execution sources "set -o pipefail"
before the command) so curl failures cause a non-zero exit.