🤖 ci: use smaller llama3.2:3b model for Ollama integration tests (#711)

Switches the Ollama model used in integration tests from `gpt-oss:20b`
(approx 10-20GB) to `llama3.2:3b` (approx 2GB).

This significantly reduces the time and resources required to download
and load the model in CI, while still providing sufficient capability
for the tool usage tests (bash, file operations).

_Generated with `mux`_

Author: ammar-agent

.github/workflows/ci.yml

@@ -134,7 +134,7 @@ jobs:
       - name: Run all integration tests with coverage
         # TEST_OLLAMA=1 enables Ollama-specific tests (now included with all integration tests)
         # --silent suppresses per-test output (17+ test files × workers = overwhelming logs)
-        run: TEST_INTEGRATION=1 bun x jest --coverage --maxWorkers=100% --silent ${{ github.event.inputs.test_filter || 'tests' }}
+        run: TEST_INTEGRATION=1 TEST_OLLAMA=1 bun x jest --coverage --maxWorkers=100% --silent ${{ github.event.inputs.test_filter || 'tests' }}
         env:
           OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}
           ANTHROPIC_API_KEY: ${{ secrets.ANTHROPIC_API_KEY }}

tests/ipcMain/ollama.test.ts

@@ -17,7 +17,9 @@ const describeOllama = shouldRunOllamaTests ? describe : describe.skip;
 // Tests require Ollama to be running and will pull models idempotently
 // Set TEST_OLLAMA=1 to enable these tests
 
-const OLLAMA_MODEL = "gpt-oss:20b";
+// Use a smaller model for CI to reduce resource usage and download time
+// while maintaining sufficient capability for tool calling tests
+const OLLAMA_MODEL = "llama3.2:3b";
 
 /**
  * Ensure Ollama model is available (idempotent).