Proxy health check fails with Ollama 0.1.44 - "unknown flag: --format" error

[misterjice]

Bug Report

Environment

• olol version: [run pip show olol to get exact version]
• Ollama version: 0.11.4 (latest stable release)
• OS: WSL2 Ubuntu
• Python version: 3.12

Problem

olol proxy fails health checks and cannot list models when connecting to olol server due to using the --format flag which doesn't exist in current stable Ollama release.

Error Message

2025-08-15 00:52:43,783 - ERROR - gRPC error: StatusCode.INTERNAL: Failed to list models: Error: unknown flag: --format
2025-08-15 00:52:43,784 - ERROR - Fallback health check failed: <_InactiveRpcError of RPC that terminated with:
status = StatusCode.INTERNAL
details = "Failed to list models: Error: unknown flag: --format"
2025-08-15 00:52:43,784 - ERROR - Server localhost:50051 health check failed

Steps to Reproduce

1. Install Ollama 0.11.4 (current latest from https://github.com/ollama/ollama/releases)
2. Verify ollama list --help shows no --format flag available
3. Start olol server: olol server --host localhost --port 50051 --ollama-host http://localhost:11434
   • Server starts successfully with: "Ollama gRPC server started on port 50051"
4. Start olol proxy: olol proxy --host localhost --port 8000 --servers "localhost:50051"
5. Observe the error when proxy attempts health checks

Expected Behavior

Proxy should successfully connect to server and list available models from Ollama, even with Ollama versions that don't support the --format flag.

Current Result

• Proxy shows "0 models" instead of actual available models
• Health checks fail
• Client requests fail due to inability to discover models

Additional Context

• ollama list works fine without the --format flag and shows models correctly
• ollama list --format json returns "unknown flag: --format"
• The olol server starts without issues - the problem is in the proxy health check logic

Suggested Fix

The proxy's health check/model discovery should:

1. Try ollama list --format json first
2. Fall back to parsing standard ollama list output when --format is unavailable
3. Or detect Ollama version and use appropriate commands accordingly