Remove LLM_MODEL from .env.example; add vendor-level base URL fallback for per-model-URL providers

[Copilot]

LLM_MODEL in .env.example is redundant (default is in code, overridable via Settings UI). Per-model-URL providers (Qwen, Moonshot, DeepSeek, etc.) only support per-model env vars like MOONSHOT_KIMI_K2_5_BASE_URL with no vendor-level fallback, which doesn't match common practice.

Changes

• .env.example: Remove LLM_MODEL line. Add vendor-level base URL examples (QWEN_BASE_URL, MOONSHOT_BASE_URL, etc.) with per-model URLs documented as optional overrides.

• src/lib/ai/provider.ts: getPerModelProvider() now resolves base URL with fallback:

  1. Per-model env var (MOONSHOT_KIMI_K2_5_BASE_URL)
  2. Vendor-level env var (MOONSHOT_BASE_URL)
  3. Error naming both options

• docs/getting-started/environment-variables.md: Add missing provider API keys and vendor-level base URL variables to the reference table.

• src/lib/ai/provider.test.ts: 5 unit tests covering priority, fallback, error messages, and all 6 providers.

Resolution logic

const perModelEnvVar = prefix + "_" + modelId.toUpperCase().replace(/[^A-Z0-9]/g, "_") + "_BASE_URL";
const vendorEnvVar = prefix + "_BASE_URL";
const baseURL = process.env[perModelEnvVar] || process.env[vendorEnvVar];

Setting MOONSHOT_BASE_URL once now works for all Moonshot models. Per-model URLs still take priority for advanced routing.

Original prompt

---

This section details on the original issue you should resolve

<issue_title>Environment variables: Remove useless LLM_MODEL from .env.example; add vendor-level base URL fallback for Qwen/Moonshot/etc.</issue_title>
<issue_description>### Problem Statement

1. Remove LLM_MODEL from .env.example

Current state: .env.example contains a commented line:

# LLM_MODEL=gpt-4o-mini

Problem: This is redundant. The default model is already defined in code (src/lib/ai/models.ts: DEFAULT_MODEL = process.env.LLM_MODEL || "gpt-4o-mini"), and users can override the default in the Settings UI. Listing LLM_MODEL in the example file suggests it is a required or recommended example variable, but it is not needed for a minimal setup.

Proposal: Remove the LLM_MODEL line from .env.example. Optionally keep it only in documentation (e.g. environment-variables.md) for advanced overrides. The in-repo example file should not advertise it as a standard env var to set.

---

2. Per-model base URL naming and fallback to vendor-level base URL

Current state: For providers such as Qwen, Moonshot, DeepSeek, MiniMax, and Zhipu, the app only supports per-model base URL environment variables, whose names are derived from the model id:

• Examples: QWEN_QWEN3_235B_BASE_URL, QWEN_QWEN3_5_397B_BASE_URL, MOONSHOT_KIMI_K2_5_BASE_URL, DEEPSEEK_DEEPSEEK_V3_2_BASE_URL, etc.
• Logic in src/lib/ai/provider.ts (getPerModelProvider): baseURL = process.env[prefix + "_" + modelId.toUpperCase().replace(/[^A-Z0-9]/g, "_") + "_BASE_URL"].
• If that variable is unset, the app throws and requires the user to set a per-model URL.

Problems:

1. Binding the env var name to model_id is unusual. Most ecosystems use a single vendor-level base URL (e.g. OPENAI_BASE_URL, ANTHROPIC_BASE_URL, GEMINI_BASE_URL). Users expect to set e.g. MOONSHOT_BASE_URL or QWEN_BASE_URL once and use any model from that provider.
2. No fallback: When a per-model URL is not set, the app does not try a vendor-level URL; it fails. So “per-model URL only” is the only supported pattern for these providers, which does not match common practice.

Proposal:

• Keep per-model base URLs for the “Per-model base URLs” use case (e.g. different endpoints per model).
• Add vendor-level base URL environment variables for each of these providers, e.g.:
  • QWEN_BASE_URL
  • MOONSHOT_BASE_URL
  • DEEPSEEK_BASE_URL
  • MINIMAX_BASE_URL
  • ZHIPU_BASE_URL
• Resolution order: For a given provider and model:
  1. If the per-model base URL env var is set (e.g. MOONSHOT_KIMI_K2_5_BASE_URL), use it.
  2. Else if the vendor base URL env var is set (e.g. MOONSHOT_BASE_URL), use it.
  3. Else, if the provider requires a base URL (e.g. no public default), fail with a clear error asking the user to set either the per-model or the vendor-level URL.

This way, “Per-model base URLs” remain supported for advanced cases, and in the common case where all models of a provider use the same endpoint, a single MOONSHOT_BASE_URL / QWEN_BASE_URL etc. is enough and aligns with common practice.

---

Summary

• .env.example: Remove the commented LLM_MODEL line.
• Provider base URLs: Introduce vendor-level *_BASE_URL env vars for Qwen, Moonshot, DeepSeek, MiniMax, Zhipu; use them when the corresponding per-model base URL is not set.

Proposed Solution

See Statement

Alternatives Considered

No response

Feature Area

AI / Chat / Agent

Additional Context

No response</issue_description>

<agent_instructions>Please fix the bugs reported in this issue.</agent_instructions>

Comments on the Issue (you are @copilot in this section)

• Fixes Environment variables: Remove useless LLM_MODEL from .env.example; add vendor-level base URL fallback for Qwen/Moonshot/etc. #15

---

📍 Connect Copilot coding agent with Jira, Azure Boards or Linear to delegate work to Copilot in one click without leaving your project management tool.

[Copilot]

Pull request overview

This PR streamlines AI provider configuration by removing the redundant LLM_MODEL example from .env.example and adding a vendor-level *_BASE_URL fallback for “per-model base URL” providers (Qwen/Moonshot/DeepSeek/MiniMax/Zhipu/SH-Lab), aligning their behavior with the rest of the provider ecosystem.

Changes:

• Removed LLM_MODEL from .env.example and documented vendor-level *_BASE_URL variables with optional per-model overrides.
• Updated getPerModelProvider() to resolve base URL via per-model env var → vendor-level env var → explicit error listing both.
• Added unit tests verifying priority/fallback/error behavior across all per-model-URL providers, and expanded env var documentation.

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated no comments.

File	Description
src/lib/ai/provider.ts	Implements vendor-level base URL fallback and clearer error messaging for per-model-URL providers.
src/lib/ai/provider.test.ts	Adds Vitest coverage for per-model vs vendor base URL resolution and error messaging.
docs/getting-started/environment-variables.md	Documents missing provider API keys and vendor-level base URL variables.
.env.example	Removes LLM_MODEL example and adds vendor-level base URL examples + per-model override notes.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.