ollama tool calling

[konsumer]

I am using ghcr.io/servicestack/llms:latest and ollama 0.15.2. When try to use tools, it calls them, but doesn't seem to handle the output, regardless of model.

(it called it, but stops there)

This seems to work fine with Gemini, Claude, and a few off HuggingFace.

Do I need some particular model, or is it a problem with ollama/llms?

[mythz]

Does the debug log reveal any errors?

DEBUG=1 llms --serve 8000 --verbose

[konsumer]

Not really. A lot of SQL logs, but no errors.

Additionally, I can keep talking to it, but it seems like it just doesn't see the output of tools. When pressed, it just hallucinates (seems based on the tools I have enabled, like calc and plex.)

Which is just made up. This is output from claude haiku on same llms instance, which is correct (I have some code mounted in PWD):

The current directory contains several subdirectories:

1. /null0
2. /aitorrent
3. /raycart
4. /REDACTED
5. /REDACTED
6. /fastboot
7. /null0-api

Each directory appears to be 4096 bytes in size, which is typical for directory entries. Would you like me to explore the contents of any of these specific directories?

And I asked about null0

This looks like a software project directory. Here are some notable contents:

• Markdown files: README.md, API.md, THANKS.md
• Configuration files: .prettierrc, .clang-format, CMakeLists.txt
• Node.js related: package.json, package-lock.json, node_modules
• Directories:
  • /webroot
  • /host
  • /node_modules
  • /tools
  • /build
  • /wbuild
  • /carts
  • /api
• Version control: .git, .gitignore
• License file: LICENSE

Would you like me to examine the contents of any of these files or directories in more detail?

[konsumer]

Here is ollama minmax-m2:cloud which should also work with tools. I tried enabling only list_directory, glob_paths, and run_python, to minimize confusion. It had the same issue, here is the log:

llms-1  | POST http://ollama:11434/v1/chat/completions
llms-1  | {
llms-1  |   "model": "minimax-m2:cloud",
llms-1  |   "messages": [
llms-1  |     {
llms-1  |       "role": "system",
llms-1  |       "content": "You have access to specialized skills that extend your capabilities with domain-specific knowledge, workflows, and tools. \nSkills are modular packages containing instructions, scripts, references, and assets for particular tasks.\n\n## Using Skills\n\nUse the skill tool to read a skill's main instructions and guidance, e.g:\n```\nskill({ name: \"skill-name\" })\n```\n\n\nTo read a specific file within a skill (scripts, references, assets):\n```\nskill({ name: \"skill-name\", file: \"relative/path/to/file\" })\n```\n\n\nExamples:\n- `skill({ name: \"create-plan\" })` - Read the create-plan skill's SKILL.md instructions\n- `skill({ name: \"web-artifacts-builder\", file: \"scripts/init-artifact.sh\" })` - Read a specific script\n\n## When to Use Skills\n\nYou should read the appropriate skill BEFORE starting work on relevant tasks. Skills contain best practices, scripts, and reference materials that significantly improve output quality.\n\n**Skill Selection Guidelines:**\n- Match the task to available skill descriptions\n- Multiple skills may be relevant - read all that apply\n- Read the skill first, then follow its instructions\n\n## Available Skills\n<available_skills>\n <skill>\n  <name>create-plan</name>\n  <description>Create a concise plan. Use when a user explicitly asks for a plan related to a coding task.</description>\n  <location>/home/llms/.llms/.agent/skills/create-plan</location>\n </skill>\n</available_skills>\n\n## Important Notes\n\n- Always read the skill BEFORE starting implementation\n- Skills may contain scripts that can be executed directly without loading into context\n- Multiple skills can and should be combined when tasks span multiple domains\n- If a skill references additional files (references/, scripts/, assets/), read those as needed during execution",
llms-1  |       "timestamp": 1769740879319
llms-1  |     },
llms-1  |     {
llms-1  |       "timestamp": 1769740879318,
llms-1  |       "role": "user",
llms-1  |       "content": [
llms-1  |         {
llms-1  |           "type": "text",
llms-1  |           "text": "I am testing tools. can you list the current directory"
llms-1  |         }
llms-1  |       ]
llms-1  |     }
llms-1  |   ],
llms-1  |   "modalities": [
llms-1  |     "text"
llms-1  |   ],
llms-1  |   "systemPrompt": "You have access to specialized skills that extend your capabilities with domain-specific knowledge, workflows, and tools. \nSkills are modular packages containing instructions, scripts, references, and assets for particular tasks.\n\n## Using Skills\n\nUse the skill tool to read a skill's main instructions and guidance, e.g:\n```\nskill({ name: \"skill-name\" })\n```\n\n\nTo read a specific file within a skill (scripts, references, assets):\n```\nskill({ name: \"skill-name\", file: \"relative/path/to/file\" })\n```\n\n\nExamples:\n- `skill({ name: \"create-plan\" })` - Read the create-plan skill's SKILL.md instructions\n- `skill({ name: \"web-artifacts-builder\", file: \"scripts/init-artifact.sh\" })` - Read a specific script\n\n## When to Use Skills\n\nYou should read the appropriate skill BEFORE starting work on relevant tasks. Skills contain best practices, scripts, and reference materials that significantly improve output quality.\n\n**Skill Selection Guidelines:**\n- Match the task to available skill descriptions\n- Multiple skills may be relevant - read all that apply\n- Read the skill first, then follow its instructions\n\n## Available Skills\n<available_skills>\n <skill>\n  <name>create-plan</name>\n  <description>Create a concise plan. Use when a user explicitly asks for a plan related to a coding task.</description>\n  <location>/home/llms/.llms/.agent/skills/create-plan</location>\n </skill>\n</available_skills>\n\n## Important Notes\n\n- Always read the skill BEFORE starting implementation\n- Skills may contain scripts that can be executed directly without loading into context\n- Multiple skills can and should be combined when tasks span multiple domains\n- If a skill references additional files (references/, scripts/, assets/), read those as needed during execution",
llms-1  |   "tools": [
llms-1  |     {
llms-1  |       "type": "function",
llms-1  |       "function": {
llms-1  |         "name": "list_directory",
llms-1  |         "description": "List directory contents",
llms-1  |         "parameters": {
llms-1  |           "type": "object",
llms-1  |           "properties": {
llms-1  |             "path": {
llms-1  |               "type": "string"
llms-1  |             }
llms-1  |           },
llms-1  |           "required": [
llms-1  |             "path"
llms-1  |           ]
llms-1  |         }
llms-1  |       }
llms-1  |     },
llms-1  |     {
llms-1  |       "type": "function",
llms-1  |       "function": {
llms-1  |         "name": "glob_paths",
llms-1  |         "description": "Find files and directories matching a glob pattern",
llms-1  |         "parameters": {
llms-1  |           "type": "object",
llms-1  |           "properties": {
llms-1  |             "pattern": {
llms-1  |               "type": "string"
llms-1  |             },
llms-1  |             "extensions": {
llms-1  |               "type": "array",
llms-1  |               "items": {
llms-1  |                 "type": "string"
llms-1  |               }
llms-1  |             },
llms-1  |             "sort_by": {
llms-1  |               "type": "string"
llms-1  |             },
llms-1  |             "max_results": {
llms-1  |               "type": "integer"
llms-1  |             }
llms-1  |           },
llms-1  |           "required": [
llms-1  |             "pattern"
llms-1  |           ]
llms-1  |         }
llms-1  |       }
llms-1  |     },
llms-1  |     {
llms-1  |       "type": "function",
llms-1  |       "function": {
llms-1  |         "name": "run_python",
llms-1  |         "description": "Execute Python code in a temporary sandboxed environment.\n    Uses ulimit for resource restriction and runs in a temporary directory.",
llms-1  |         "parameters": {
llms-1  |           "type": "object",
llms-1  |           "properties": {
llms-1  |             "code": {
llms-1  |               "type": "string"
llms-1  |             }
llms-1  |           },
llms-1  |           "required": [
llms-1  |             "code"
llms-1  |           ]
llms-1  |         }
llms-1  |       }
llms-1  |     }
llms-1  |   ],
llms-1  |   "metadata": {
llms-1  |     "tools": "list_directory,glob_paths,run_python",
llms-1  |     "skills": "all"
llms-1  |   },
llms-1  |   "stream": false
llms-1  | }
llms-1  | DEBUG [app]: SQL> SELECT * FROM thread WHERE user IS NULL AND id = :id
llms-1  | {'id': '21'}
llms-1  | DEBUG [app]: get_thread_updates: 21 / 2026-01-29 18:41:19.321000 < 2026-01-29 18:41:19.320300 / False
llms-1  | DEBUG [app]: SQL> SELECT * FROM thread WHERE user IS NULL AND id = :id
llms-1  | {'id': 21}
llms-1  | DEBUG [app]: SQL> SELECT * FROM thread WHERE user IS NULL AND id = :id
llms-1  | {'id': '21'}
llms-1  | DEBUG [app]: SQL> SELECT updatedAt FROM thread WHERE user IS NULL AND id = :id
llms-1  | {'id': 21}
llms-1  | DEBUG [app]: create_response
llms-1  | DEBUG [app]: SQL> INSERT INTO request (model, duration, cost, inputPrice, inputTokens, inputCachedTokens, outputPrice, outputTokens, finishReason, provider, providerModel, providerRef, threadId, title, startedAt, totalTokens, usage, completedAt, ref, createdAt, updatedAt) VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?)
llms-1  | ('minimax-m2 cloud', 0, 0.0, 0, 653, 0, 0, 76, 'tool_calls', 'ollama', 'minimax-m2:cloud', None, '21', 'I am testing tools. can you list the current directory', datetime.datetime(2026, 1, 29, 18, 41, 19, 320956), 729, '{"prompt_tokens": 653, "completion_tokens": 76, "total_tokens": 729, "duration": 0}', datetime.datetime(2026, 1, 29, 18, 41, 20, 282086), 'chatcmpl-951', datetime.datetime(2026, 1, 29, 18, 41, 20, 282142), datetime.datetime(2026, 1, 29, 18, 41, 20, 282142))
llms-1  | DEBUG [app]: lastrowid 60, rowcount 1
llms-1  | DEBUG [app]: SQL> UPDATE thread SET model = :model, providerModel = :providerModel, modelInfo = :modelInfo, messages = :messages, tools = :tools, completedAt = :completedAt, providerResponse = :providerResponse, updatedAt = :updatedAt WHERE id = :id
llms-1  | {'model': 'minimax-m2 cloud', 'providerModel': 'minimax-m2:cloud', 'modelInfo': '{"id": "minimax-m2:cloud", "name": "minimax-m2 cloud", "modalities": {"input": ["text"], "output": ["text"]}, "cost": {"input": 0, "output": 0}}', 'messages': '[{"role": "system", "content": "You have access to specialized skills that extend your capabilities with domain-specific knowledge, workflows, and tools. \\nSkills are modular packages containing instructions, scripts, references, and assets for particular tasks.\\n\\n## Using Skills\\n\\nUse the skill tool to read a skill\'s main instructions and guidance, e.g:\\n```\\nskill({ name: \\"skill-name\\" })\\n```\\n\\n\\nTo read a specific file within a skill (scripts, references, assets):\\n```\\nskill({ name: \\"skill-name\\", file: \\"relative/path/to/file\\" })\\n```\\n\\n\\nExamples:\\n- `skill({ name: \\"create-plan\\" })` - Read the create-plan skill\'s SKILL.md instructions\\n- `skill({ name: \\"web-artifacts-builder\\", file: \\"scripts/init-artifact.sh\\" })` - Read a specific script\\n\\n## When to Use Skills\\n\\nYou should read the appropriate skill BEFORE starting work on relevant tasks. Skills contain best practices, scripts, and reference materials that significantly improve output quality.\\n\\n**Skill Selection Guidelines:**\\n- Match the task to available skill descriptions\\n- Multiple skills may be relevant - read all that apply\\n- Read the skill first, then follow its instructions\\n\\n## Available Skills\\n<available_skills>\\n <skill>\\n  <name>create-plan</name>\\n  <description>Create a concise plan. Use when a user explicitly asks for a plan related to a coding task.</description>\\n  <location>/home/llms/.llms/.agent/skills/create-plan</location>\\n </skill>\\n</available_skills>\\n\\n## Important Notes\\n\\n- Always read the skill BEFORE starting implementation\\n- Skills may contain scripts that can be executed directly without loading into context\\n- Multiple skills can and should be combined when tasks span multiple domains\\n- If a skill references additional files (references/, scripts/, assets/), read those as needed during execution", "timestamp": 1769740879319}, {"timestamp": 1769740879318, "role": "user", "content": [{"type": "text", "text": "I am testing tools. can you list the current directory"}], "model": "minimax-m2 cloud", "usage": {"tokens": 653, "price": 0, "cost": 0.0}}, {"role": "assistant", "content": "I\'ll list the current directory contents for you using the `list_directory` tool.", "reasoning": "The user is asking me to list the current directory. I can use the `list_directory` tool for this purpose. This is a straightforward request to test the toolsfunctionality.", "tool_calls": [{"id": "call_2xlunmzy", "index": 0, "type": "function", "function": {"name": "list_directory", "arguments": "{\\"path\\":\\".\\"}"}}], "timestamp": 1769740880282, "model": "minimax-m2 cloud", "usage": {"tokens": 76, "price": 0, "cost": 0.0, "duration": 0}}]', 'tools': '[{"type": "function", "function": {"name": "list_directory", "description": "List directory contents", "parameters": {"type": "object", "properties": {"path": {"type": "string"}}, "required": ["path"]}}}, {"type": "function", "function": {"name": "glob_paths", "description": "Find files and directories matching a glob pattern", "parameters": {"type": "object", "properties": {"pattern": {"type": "string"}, "extensions": {"type": "array", "items": {"type": "string"}}, "sort_by": {"type": "string"}, "max_results": {"type": "integer"}}, "required": ["pattern"]}}}, {"type": "function", "function": {"name": "run_python", "description": "Execute Python code in a temporary sandboxed environment.\\n    Uses ulimit for resource restriction and runs in a temporary directory.", "parameters": {"type": "object", "properties": {"code": {"type": "string"}}, "required": ["code"]}}}]', 'completedAt': datetime.datetime(2026, 1, 29, 18, 41, 20, 282086), 'providerResponse': '{"id": "chatcmpl-951", "object": "chat.completion", "created": 1769740880, "model": "minimax-m2:cloud", "system_fingerprint": "fp_ollama", "choices": [{"index": 0, "message": {"role": "assistant", "content": "I\'ll list the current directory contents for you using the `list_directory` tool.", "reasoning": "The user is asking me to list the current directory. I can use the `list_directory` tool for this purpose. This is a straightforward request to test the toolsfunctionality.", "tool_calls": [{"id": "call_2xlunmzy", "index": 0, "type": "function", "function": {"name": "list_directory", "arguments": "{\\"path\\":\\".\\"}"}}], "timestamp": 1769740880282, "model": "minimax-m2 cloud", "usage": {"tokens": 76, "price": 0, "cost": 0.0, "duration": 0}}, "finish_reason": "tool_calls"}], "usage": {"prompt_tokens": 653, "completion_tokens": 76, "total_tokens": 729, "duration": 0}, "metadata": {"duration": 960, "pricing": "0/0"}}', 'updatedAt': datetime.datetime(2026, 1, 29, 18, 41, 20, 282394), 'id': '21'}
llms-1  | DEBUG [app]: lastrowid 60, rowcount 1
llms-1  | DEBUG [app]: SQL> SELECT id, user, threadId, createdAt, updatedAt, title, model, duration, cost, inputPrice, inputTokens, inputCachedTokens, outputPrice, outputTokens, totalTokens, usage, provider, providerModel, providerRef, finishReason, startedAt, completedAt, error, stackTrace, ref FROM request WHERE user IS NULL AND threadId = :threadId ORDER BY id DESC LIMIT :take OFFSET :skip
llms-1  | {'take': 50, 'skip': 0, 'threadId': '21'}
llms-1  | DEBUG: {
llms-1  |   "id": "chatcmpl-951",
llms-1  |   "object": "chat.completion",
llms-1  |   "created": 1769740880,
llms-1  |   "model": "minimax-m2:cloud",
llms-1  |   "system_fingerprint": "fp_ollama",
llms-1  |   "choices": [
llms-1  |     {
llms-1  |       "index": 0,
llms-1  |       "message": {
llms-1  |         "role": "assistant",
llms-1  |         "content": "I'll list the current directory contents for you using the `list_directory` tool.",
llms-1  |         "reasoning": "The user is asking me to list the current directory. I can use the `list_directory` tool for this purpose. This is a straightforward request to test the toolsfunctionality.",
llms-1  |         "tool_calls": [
llms-1  |           {
llms-1  |             "id": "call_2xlunmzy",
llms-1  |             "index": 0,
llms-1  |             "type": "function",
llms-1  |             "function": {
llms-1  |               "name": "list_directory",
llms-1  |               "arguments": "{\"path\":\".\"}"
llms-1  |             }
llms-1  |           }
llms-1  |         ],
llms-1  |         "timestamp": 1769740880282,
llms-1  |         "model": "minimax-m2 cloud",
llms-1  |         "usage": {
llms-1  |           "tokens": 76,
llms-1  |           "price": 0,
llms-1  |           "cost": 0.0,
llms-1  |           "duration": 0
llms-1  |         }
llms-1  |       },
llms-1  |       "finish_reason": "tool_calls"
llms-1  |     }
llms-1  |   ],
llms-1  |   "usage": {
llms-1  |     "prompt_tokens": 653,
llms-1  |     "completion_tokens": 76,
llms-1  |     "total_tokens": 729,
llms-1  |     "duration": 0
llms-1  |   },
llms-1  |   "metadata": {
llms-1  |     "duration": 960,
llms-1  |     "pricing": "0/0"
llms-1  |   }
llms-1  | }
llms-1  | DEBUG [app]: SQL> UPDATE thread SET inputTokens = :inputTokens, outputTokens = :outputTokens, cost = :cost, stats = :stats, updatedAt = :updatedAt WHERE id = :id
llms-1  | {'inputTokens': 653, 'outputTokens': 76, 'cost': 0, 'stats': '{"inputTokens": 653, "outputTokens": 76, "cost": 0, "duration": 0, "requests": 1}', 'updatedAt': datetime.datetime(2026, 1, 29, 18, 41, 20, 282868), 'id': '21'}
llms-1  | DEBUG [app]: lastrowid 60, rowcount 1
llms-1  | DEBUG [app]: SQL> SELECT updatedAt FROM thread WHERE user IS NULL AND id = :id
llms-1  | {'id': 21}
llms-1  | DEBUG [app]: get_thread_updates: 21 / 2026-01-29 18:41:20.282868 < 2026-01-29 18:41:19.321000 / False
llms-1  | DEBUG [app]: SQL> SELECT * FROM thread WHERE user IS NULL AND id = :id
llms-1  | {'id': 21}

[mythz]

The issue was because the models returned by ollama model resolution wasn't marked as supporting tool calling which should be resolved in this commit: d6dfee4

This change is now available from v3.0.23+

[konsumer]

looks like the docker is still at llms v3.0.16. Any idea when it will update?

[mythz]

It should be automatically updated after every release, I've just checked it and it is v3.0.23. Did you update?

docker pull ghcr.io/servicestack/llms:latest

[konsumer]

Ah, yes, I am using a FROM in a custom Dockerfile that extends & adds all the deps for my tools and stuff, and I forgot to build. works perfect!

Thanks! this is sick.

Here is a (free and local!) test with my torrent/media MCP