fix: honor LLM_BASE_URL / EMBEDDING_BASE_URL env-var overrides in bootstrap-local

[djravine]

Closes #16.

Summary

scripts/dev/bootstrap-local.mjs writes hardcoded LLM_BASE_URL / EMBEDDING_BASE_URL values to .env.local regardless of any process.env overrides — ensureEnvValue(key, value) never consults the environment.

This is a latent bug on main today (nothing currently sets those env vars at install time), but the in-flight #15 (one-line install scripts) is about to surface it: its --llm-port / --embedding-port flags propagate into the reachability check but get silently overwritten downstream when bootstrap-local lands its hardcoded defaults.

Change

Four lines in bootstrap-local.mjs — wrap the LLM_BASE_URL / EMBEDDING_BASE_URL calls in both the qwen-local and llama-cpp branches with process.env.X || '<default>':

ensureEnvValue('LLM_BASE_URL', process.env.LLM_BASE_URL || 'http://127.0.0.1:11434/v1');  // qwen-local
ensureEnvValue('EMBEDDING_BASE_URL', process.env.EMBEDDING_BASE_URL || 'http://127.0.0.1:11434/v1');
ensureEnvValue('LLM_BASE_URL', process.env.LLM_BASE_URL || 'http://127.0.0.1:8080/v1');   // llama-cpp
ensureEnvValue('EMBEDDING_BASE_URL', process.env.EMBEDDING_BASE_URL || 'http://127.0.0.1:8081/v1');

The openai branch needs no change — it doesn't write base URLs (OpenAI uses the SDK default).

The qwen-local branch is updated for defensive symmetry. Nothing sets those env vars at install time today, but mirroring the override pattern future-proofs against e.g. pointing Ollama at a non-default host and keeps the behaviour consistent across local-LLM profiles.

Verification

Ran end-to-end in a clean ubuntu:24.04 container against this branch's working tree:

== Resulting .env.local (BASE_URL lines) ==
LLM_BASE_URL=http://127.0.0.1:9090/v1
EMBEDDING_BASE_URL=http://127.0.0.1:9091/v1

== Assertions ==
  PASS: both BASE_URL env vars honored

Sequence: apt bootstrap → nvm + Node 20 → rsync working tree → npm install → LLM_BASE_URL=http://127.0.0.1:9090/v1 EMBEDDING_BASE_URL=http://127.0.0.1:9091/v1 npm run setup:local -- --profile llama-cpp → grep .env.local for the BASE_URL lines.

Without the fix the same flow produces the hardcoded 8080/8081 defaults — confirmed earlier when the bug was originally identified.

Follow-up

Once this merges, the llama-cpp-with-mock-servers scenario in scripts/install-tests/linux-edge-cases.sh (introduced by #15) can be extended to additionally assert that .env.local ends up with the custom URLs — making it a full end-to-end regression guard for both the install.sh env-var passing and this bootstrap-local override path. The scenario currently ships with an inline LIMITATION: comment pointing at #16 explaining the deferred assertion.

🤖 Generated with Claude Code