fix: respect contextLength from YAML model config

[RomneyDa]

Summary

Fixes contextLength not being respected when set in YAML config, addressing a long-standing issue reported by multiple users.

• Add contextLength to YAML model schema at the model level — previously it was only recognized inside defaultCompletionOptions, and zod silently stripped it if placed at the model level (where JSON config supports it)
• Read contextLength from both locations — model-level takes precedence over defaultCompletionOptions.contextLength
• Fix GUI selector fallback — selectSelectedChatModelContextLength was falling back to DEFAULT_MAX_TOKENS (4096) instead of DEFAULT_CONTEXT_LENGTH (32768)

Both of these YAML configs now work:

models:
  - name: my-model
    provider: openai
    model: my-model
    contextLength: 65536
    defaultCompletionOptions:
      maxTokens: 32000

models:
  - name: my-model
    provider: openai
    model: my-model
    defaultCompletionOptions:
      contextLength: 65536
      maxTokens: 32000

Closes #4638
Closes #4304

Test plan

• Set contextLength at model level in YAML config — verify it's respected (no truncation errors)
• Set contextLength inside defaultCompletionOptions — verify it still works
• Set contextLength at both levels — verify model-level takes precedence
• Verify validation warning fires when maxTokens is too close to contextLength at either level

---

Summary by cubic

Respect contextLength from YAML configs. Model-level contextLength is supported and takes precedence; GUI fallback, validation, and docs now match.

• Bug Fixes

  • Added contextLength to the YAML model schema at the model level in @continuedev/config-yaml, matching JSON config.
  • Read contextLength from the model level or defaultCompletionOptions, with model-level taking precedence.
  • Fixed GUI selector fallback to use DEFAULT_CONTEXT_LENGTH instead of DEFAULT_MAX_TOKENS.
  • Validation now uses the effective contextLength and maxTokens, warning when they’re too close.
  • Docs: YAML examples use defaultCompletionOptions.contextLength; JSON examples show model-level contextLength. Fixed Ollama guide key name to defaultCompletionOptions.

• Refactors

  • Replaced find with includes for capability checks in modelConfigToBaseLLM.
  • Extracted env-to-options mapping into applyEnvOptions to reduce complexity.

^{Written for commit c254c9f. Summary will update on new commits.}

[cubic-dev-ai]

No issues found across 4 files

[continue]

Test Coverage Review

This PR fixes how contextLength is respected from YAML model config. While the changes are straightforward bug fixes, there's one test gap worth noting:

Suggested Test Addition

core/config/yaml/models.vitest.ts - Consider adding a test case to verify that model.contextLength takes precedence over model.defaultCompletionOptions?.contextLength:

it("should prioritize model.contextLength over defaultCompletionOptions.contextLength", async () => {
  const model: ModelConfig = {
    name: "test-model",
    provider: "openai",
    model: "gpt-4",
    contextLength: 128000, // Should take precedence
    defaultCompletionOptions: {
      contextLength: 64000, // Should be ignored
    },
  };

  const result = await llmsFromModelConfig({
    model,
    uniqueId: "test-id",
    llmLogger: mockLLMLogger,
    config: mockConfig,
  });

  expect(result).toHaveLength(1);
  expect(result[0].contextLength).toBe(128000);
});

it("should fall back to defaultCompletionOptions.contextLength when model.contextLength is not set", async () => {
  const model: ModelConfig = {
    name: "test-model",
    provider: "openai",
    model: "gpt-4",
    defaultCompletionOptions: {
      contextLength: 64000,
    },
  };

  const result = await llmsFromModelConfig({
    model,
    uniqueId: "test-id",
    llmLogger: mockLLMLogger,
    config: mockConfig,
  });

  expect(result).toHaveLength(1);
  expect(result[0].contextLength).toBe(64000);
});

This would ensure the prioritization logic (model.contextLength ?? model.defaultCompletionOptions?.contextLength) is explicitly covered.

No Tests Needed For

• Schema change (contextLength field addition) - Low risk, validated by zod schema
• GUI constant fix (DEFAULT_MAX_TOKENS → DEFAULT_CONTEXT_LENGTH) - Clear bug fix, correct constant
• Validation logic - Pre-existing gap (no tests for validateConfigYaml), not introduced by this PR