Add --no-stream flag for reasoning models

[gwpl]

AI Assistant:

Summary

This PR adds a minimal --no-stream CLI flag to disable response streaming, resolving issues with reasoning models (o1, o3, o3-pro) that require organization verification for streaming or don't support streaming on certain platforms (Azure OpenAI).

Fixes #430

Problem

Currently, mods hardcodes streaming for all API calls, which causes 400 errors for:

• o3/o3-mini/o3-pro models without org verification: "param": "stream", "code": "unsupported_value"
• o1 models on Azure OpenAI (streaming not supported)
• Any reasoning model with streaming restrictions

Solution

Add --no-stream flag that:

• Disables streaming when set
• Uses regular Chat Completion API instead of streaming API
• Maintains backward compatibility (streaming enabled by default)
• Implements NonStreamingWrapper to provide consistent stream.Stream interface

Changes

1. config.go (2 changes)

• Add NoStream bool field to Config struct (line 148)
• Add help text for --no-stream flag (line 47)

2. main.go (1 change)

• Register --no-stream CLI flag (line 270)

3. internal/proto/proto.go (1 change)

• Add Stream *bool field to Request struct (line 78)

4. mods.go (1 change)

• Pass NoStream config to request (lines 440-443)

5. internal/openai/openai.go (2 changes)

• Implement conditional streaming logic (lines 96-106)
• Add NonStreamingWrapper struct implementing stream.Stream (lines 197-254)

Testing

Basic functionality tests

# Non-streaming mode works
echo "what is 2+2?" | ./mods -m gpt-4o-mini --no-stream
# Output: 2 + 2 equals 4.

# Streaming mode still works (default)
echo "what is 3+3?" | ./mods -m gpt-4o-mini
# Output: 3 + 3 equals 6.

# Help text includes new flag
./mods --help | grep no-stream
# Output: --no-stream  Disable streaming of responses (useful for reasoning models without streaming support)

Real-world o3 model test (proves the fix works)

# WITHOUT --no-stream: FAILS with streaming error
echo "test" | ./mods -m o3
# ERROR: 400 Bad Request
# "message": "Your organization must be verified to stream this model..."
# "param": "stream"
# "code": "unsupported_value"

# WITH --no-stream: WORKS!
echo "test" | ./mods -m o3 --no-stream
# Output: Hello! It looks like you're testing the chat. How can I help you today?
# Conversation saved: 6f469bf test

This proves the --no-stream flag successfully resolves issue #430 for o3/o3-pro reasoning models.

Scope & Future Work

This PR implements ONLY the CLI flag. Per-model configuration in config_template.yml is intentionally left for future work to keep this change minimal and focused on resolving the immediate issue.

Future enhancements could include:

• Per-model stream: false configuration in config files
• Auto-detection of models that don't support streaming
• Azure OpenAI specific handling

Backward Compatibility

✅ Fully backward compatible:

• Default behavior unchanged (streaming enabled)
• Existing scripts/workflows continue to work
• New flag is opt-in only

🤖 Generated with Claude Code

[gwpl]

AI Assistant:

CI Lint Status Update

The macOS lint failure is due to a pre-existing issue in the upstream codebase, not introduced by this PR.

Error: os/exec.Command must not be called. use os/exec.CommandContext (noctx) at mods.go:485

Analysis:

• This error is at line 485 of mods.go
• The --no-stream implementation only modifies lines 440-443 of mods.go
• Line 485 is unrelated to the streaming changes
• This noctx violation exists in the base branch

Files actually modified by this PR:

1. config.go (lines 47, 148)
2. main.go (line 270)
3. internal/proto/proto.go (line 78)
4. mods.go (lines 440-443)
5. internal/openai/openai.go (lines 96-254)

Build Status: ✅ All build checks pass on ubuntu, macos, and windows

Testing: Successfully tested with o3 model:

• WITHOUT --no-stream: 400 error (streaming not allowed)
• WITH --no-stream: Works successfully ✅

Request: The maintainers may want to address the pre-existing noctx issue separately, or accept this PR despite the lint warning since it's unrelated to the changes introduced here.