harn-vm 0.8.129

# ---------- NVIDIA NIM — OpenAI-compatible API Catalog -----------------------
# NVIDIA NIM exposes these routes on /v1/chat/completions. Keep family
# semantics explicit because the provider mixes Nemotron, DeepSeek, MiniMax,
# Kimi, Mistral, GPT-OSS, GLM, and Step models behind one OpenAI-compatible
# endpoint.

[[provider.nvidia]]
model_match = "*gpt-oss-*"
native_tools = true
preferred_tool_format = "native"
structured_output = "native"
thinking_modes = ["effort"]
reasoning_effort_supported = true
reasoning_effort_levels = ["low", "medium", "high"]
reasoning_required_for_tools = true
text_tool_wire_format_supported = true
prefers_xml_scaffolding = false
prefers_markdown_scaffolding = true
structured_output_mode = "native_json"
supports_assistant_prefill = false
prefers_role_developer = false
prefers_xml_tools = false
thinking_block_style = "reasoning_summary"

[[provider.nvidia]]
model_match = "*nemotron-3-nano-omni*"
native_tools = true
preferred_tool_format = "native"
structured_output = "native"
thinking_modes = ["enabled"]
vision = true
vision_supported = true
video = true
audio = true
text_tool_wire_format_supported = true
prefers_xml_scaffolding = false
prefers_markdown_scaffolding = true
structured_output_mode = "native_json"
supports_assistant_prefill = false
prefers_role_developer = false
prefers_xml_tools = false
thinking_block_style = "inline"

[[provider.nvidia]]
model_match = "*nemotron-3-nano-30b-a3b*"
native_tools = true
preferred_tool_format = "native"
tool_mode_parity = "interchangeable"
tool_mode_parity_notes = "2026-06-20 Harn agent-loop smoke: NVIDIA NIM Nemotron 3 Nano completed both native and JSON tool loops with reasoning disabled. Earlier native false negatives were caused by Harn's parser treating terse final answers as billed no-ops; keep native preferred."
structured_output = "native"
thinking_modes = ["enabled"]
auto_reasoning_overrides = { agent = "off", verify = "off", code = "off" }
text_tool_wire_format_supported = true
prefers_xml_scaffolding = false
prefers_markdown_scaffolding = true
structured_output_mode = "native_json"
supports_assistant_prefill = false
prefers_role_developer = false
prefers_xml_tools = false
thinking_block_style = "inline"

[[provider.nvidia]]
model_match = "*nemotron-3*"
native_tools = true
preferred_tool_format = "native"
structured_output = "native"
thinking_modes = ["enabled"]
text_tool_wire_format_supported = true
prefers_xml_scaffolding = false
prefers_markdown_scaffolding = true
structured_output_mode = "native_json"
supports_assistant_prefill = false
prefers_role_developer = false
prefers_xml_tools = false
thinking_block_style = "inline"

[[provider.nvidia]]
model_match = "*deepseek-v4*"
native_tools = true
preferred_tool_format = "native"
structured_output = "native"
thinking_modes = ["enabled"]
prompt_caching = true
text_tool_wire_format_supported = true
prefers_xml_scaffolding = false
prefers_markdown_scaffolding = true
structured_output_mode = "native_json"
supports_assistant_prefill = false
prefers_role_developer = false
prefers_xml_tools = false
thinking_block_style = "inline"

[[provider.nvidia]]
model_match = "*minimax-m3*"
native_tools = true
preferred_tool_format = "native"
structured_output = "delimited"
thinking_modes = ["adaptive"]
vision = true
vision_supported = true
video = true
prompt_caching = true
text_tool_wire_format_supported = true
prefers_xml_scaffolding = false
prefers_markdown_scaffolding = true
structured_output_mode = "delimited"
supports_assistant_prefill = false
prefers_role_developer = false
prefers_xml_tools = false
thinking_block_style = "inline"

[[provider.nvidia]]
model_match = "*minimax-m2.7*"
native_tools = true
preferred_tool_format = "native"
structured_output = "delimited"
thinking_modes = ["enabled"]
prompt_caching = true
text_tool_wire_format_supported = true
prefers_xml_scaffolding = false
prefers_markdown_scaffolding = true
structured_output_mode = "delimited"
supports_assistant_prefill = false
prefers_role_developer = false
prefers_xml_tools = false
thinking_block_style = "inline"

[[provider.nvidia]]
model_match = "*kimi-k2.6*"
native_tools = true
preferred_tool_format = "native"
structured_output = "native"
thinking_modes = ["enabled"]
prompt_caching = true
vision = true
vision_supported = true
text_tool_wire_format_supported = true
prefers_xml_scaffolding = false
prefers_markdown_scaffolding = true
structured_output_mode = "native_json"
supports_assistant_prefill = false
prefers_role_developer = false
prefers_xml_tools = false
thinking_block_style = "inline"

[[provider.nvidia]]
model_match = "*glm-5*"
native_tools = true
preferred_tool_format = "native"
structured_output = "native"
thinking_modes = ["enabled"]
prompt_caching = true
text_tool_wire_format_supported = true
prefers_xml_scaffolding = false
prefers_markdown_scaffolding = true
structured_output_mode = "native_json"
supports_assistant_prefill = false
prefers_role_developer = false
prefers_xml_tools = false
thinking_block_style = "inline"

[[provider.nvidia]]
model_match = "*step-3.7-flash*"
native_tools = true
preferred_tool_format = "native"
tool_mode_parity = "interchangeable"
prompt_caching = true
structured_output = "native"
thinking_modes = ["enabled"]
reasoning_disable_supported = false
vision = true
vision_supported = true
text_tool_wire_format_supported = true
prefers_xml_scaffolding = false
prefers_markdown_scaffolding = true
structured_output_mode = "native_json"
supports_assistant_prefill = false
prefers_role_developer = false
prefers_xml_tools = false
thinking_block_style = "inline"

[[provider.nvidia]]
model_match = "*mistral*"
native_tools = true
preferred_tool_format = "native"
structured_output = "native"
text_tool_wire_format_supported = true
prefers_xml_scaffolding = false
prefers_markdown_scaffolding = true
structured_output_mode = "native_json"
supports_assistant_prefill = false
prefers_role_developer = false
prefers_xml_tools = false
thinking_block_style = "none"

[[provider.nvidia]]
model_match = "*gemma-4*"
native_tools = true
preferred_tool_format = "native"
structured_output = "native"
thinking_modes = ["enabled"]
vision = true
vision_supported = true
text_tool_wire_format_supported = true
prefers_xml_scaffolding = false
prefers_markdown_scaffolding = true
structured_output_mode = "native_json"
supports_assistant_prefill = false
prefers_role_developer = false
prefers_xml_tools = false
thinking_block_style = "inline"

[[provider.nvidia]]
model_match = "*"
native_tools = true
preferred_tool_format = "native"
structured_output = "native"
text_tool_wire_format_supported = true
prefers_xml_scaffolding = false
prefers_markdown_scaffolding = true
structured_output_mode = "native_json"
supports_assistant_prefill = false
prefers_role_developer = false
prefers_xml_tools = false
thinking_block_style = "none"