stop_tool_calls_and_respond! breaks tool calls?

[joshforbes]

I must be doing something wrong here, I feel like I'm missing something. Anytime I add the stop_tool_calls_and_respond! to a function, the LLM responds like it can't see any tools at all.

I, [2025-07-01T21:19:25.118808 #63129]  INFO -- : request: POST https://openrouter.ai/api/v1/chat/completions
I, [2025-07-01T21:19:25.118860 #63129]  INFO -- : request: Authorization: "Bearer[REDACTED]
Content-Type: "application/json"
X-Title: "OpenRouter Ruby Client"
HTTP-Referer: "https://github.com/OlympiaAI/open_router"
I, [2025-07-01T21:19:25.118883 #63129]  INFO -- : request: {"messages":[{"role":"user","content":"Use the available tool"}],"model":"anthropic/claude-sonnet-4","max_tokens":1000,"temperature":0.0,"tools":[{"type":"function","function":{"name":"test","parameters":{"type":"object","properties":{}},"description":"Call this tool"}}]}
I, [2025-07-01T21:19:27.058307 #63129]  INFO -- : response: Status 200
I, [2025-07-01T21:19:27.058375 #63129]  INFO -- : response: date: "Tue, 01 Jul 2025 21:19:26 GMT"
content-type: "application/json"
transfer-encoding: "chunked"
connection: "keep-alive"
access-control-allow-origin: "*"
vary: "Accept-Encoding"
server: "cloudflare"
cf-ray: "9588f1073937df73-IAD"
I, [2025-07-01T21:19:27.058411 #63129]  INFO -- : response: 
         
{"id":"gen-1751404765-sW7r6Kz2503jNJWKfCxn","provider":"Google","model":"anthropic/claude-sonnet-4","object":"chat.completion","created":1751404765,"choices":[{"logprobs":null,"finish_reason":"tool_calls","native_finish_reason":"tool_calls","index":0,"message":{"role":"assistant","content":"I'll use the available tool for you.","refusal":null,"reasoning":null,"tool_calls":[{"id":"toolu_vrtx_01PjBBiSNE7ytfzQVYkWEVpR","index":0,"type":"function","function":{"name":"test","arguments":""}}]}}],"usage":{"prompt_tokens":359,"completion_tokens":44,"total_tokens":403}}
I, [2025-07-01T21:19:27.059205 #63129]  INFO -- : request: POST https://openrouter.ai/api/v1/chat/completions
I, [2025-07-01T21:19:27.059249 #63129]  INFO -- : request: Authorization: "Bearer[REDACTED]
Content-Type: "application/json"
X-Title: "OpenRouter Ruby Client"
HTTP-Referer: "https://github.com/OlympiaAI/open_router"
I, [2025-07-01T21:19:27.059273 #63129]  INFO -- : request: {"messages":[{"role":"user","content":"Use the available tool"}],"model":"anthropic/claude-sonnet-4","max_tokens":1000,"temperature":0.0}
I, [2025-07-01T21:19:29.619985 #63129]  INFO -- : response: Status 200
I, [2025-07-01T21:19:29.620248 #63129]  INFO -- : response: date: "Tue, 01 Jul 2025 21:19:27 GMT"
content-type: "application/json"
transfer-encoding: "chunked"
connection: "keep-alive"
access-control-allow-origin: "*"
vary: "Accept-Encoding"
server: "cloudflare"
cf-ray: "9588f1135f565774-IAD"
I, [2025-07-01T21:19:29.620462 #63129]  INFO -- : response: 
         
{"id":"gen-1751404767-K5LJ3ze8JEh9MPwzwbzV","provider":"Google","model":"anthropic/claude-sonnet-4","object":"chat.completion","created":1751404767,"choices":[{"logprobs":null,"finish_reason":"stop","native_finish_reason":"stop","index":0,"message":{"role":"assistant","content":"I don't see any tools available in our current conversation. Could you please clarify what tool you'd like me to use? \n\nIf you're expecting me to have access to a specific tool (like a calculator, web search, code interpreter, etc.), you may need to:\n\n1. Specify which tool you want me to use\n2. Provide the tool or enable it in your interface\n3. Give me more context about what task you'd like me to accomplish with a tool\n\nWhat would you like me to help you with?","refusal":null,"reasoning":null}}],"usage":{"prompt_tokens":11,"completion_tokens":116,"total_tokens":127}}
"I don't see any tools available in our current conversation. Could you please clarify what tool you'd like me to use? \n" +
"\n" +
"If you're expecting me to have access to a specific tool (like a calculator, web search, code interpreter, etc.), you may need to:\n" +
"\n" +
"1. Specify which tool you want me to use\n" +
"2. Provide the tool or enable it in your interface\n" +
"3. Give me more context about what task you'd like me to accomplish with a tool\n" +
"\n" +
"What would you like me to help you with?"

But if I remove that one line:

I, [2025-07-01T21:21:04.030335 #64307]  INFO -- : request: POST https://openrouter.ai/api/v1/chat/completions
I, [2025-07-01T21:21:04.030390 #64307]  INFO -- : request: Authorization: "Bearer[REDACTED]
Content-Type: "application/json"
X-Title: "OpenRouter Ruby Client"
HTTP-Referer: "https://github.com/OlympiaAI/open_router"
I, [2025-07-01T21:21:04.030414 #64307]  INFO -- : request: {"messages":[{"role":"user","content":"Use the available tool"}],"model":"anthropic/claude-sonnet-4","max_tokens":1000,"temperature":0.0,"tools":[{"type":"function","function":{"name":"test","parameters":{"type":"object","properties":{}},"description":"Call this tool"}}]}
I, [2025-07-01T21:21:05.965741 #64307]  INFO -- : response: Status 200
I, [2025-07-01T21:21:05.965827 #64307]  INFO -- : response: date: "Tue, 01 Jul 2025 21:21:05 GMT"
content-type: "application/json"
transfer-encoding: "chunked"
connection: "keep-alive"
access-control-allow-origin: "*"
vary: "Accept-Encoding"
server: "cloudflare"
cf-ray: "9588f37168a28bd6-IAD"
I, [2025-07-01T21:21:05.965858 #64307]  INFO -- : response: 
         
{"id":"gen-1751404864-d5CHXGNIcXeCd58bea5y","provider":"Google","model":"anthropic/claude-sonnet-4","object":"chat.completion","created":1751404864,"choices":[{"logprobs":null,"finish_reason":"tool_calls","native_finish_reason":"tool_calls","index":0,"message":{"role":"assistant","content":"I'll use the available tool for you.","refusal":null,"reasoning":null,"tool_calls":[{"id":"toolu_vrtx_01A9tuQudwLYUaYhQFXeuvtF","index":0,"type":"function","function":{"name":"test","arguments":""}}]}}],"usage":{"prompt_tokens":359,"completion_tokens":44,"total_tokens":403}}
I, [2025-07-01T21:21:05.966519 #64307]  INFO -- : request: POST https://openrouter.ai/api/v1/chat/completions
I, [2025-07-01T21:21:05.966545 #64307]  INFO -- : request: Authorization: "Bearer[REDACTED]
Content-Type: "application/json"
X-Title: "OpenRouter Ruby Client"
HTTP-Referer: "https://github.com/OlympiaAI/open_router"
I, [2025-07-01T21:21:05.966568 #64307]  INFO -- : request: {"messages":[{"role":"user","content":"Use the available tool"},{"role":"assistant","content":null,"tool_calls":[{"id":"c9f5986c-9d7e-4159-8743","type":"function","function":{"name":"test","arguments":"{}"}}]},{"role":"tool","tool_call_id":"c9f5986c-9d7e-4159-8743","name":"test","content":"Let the user know that the tool was used"}],"model":"anthropic/claude-sonnet-4","max_tokens":1000,"temperature":0.0,"tools":[{"type":"function","function":{"name":"test","parameters":{"type":"object","properties":{}},"description":"Call this tool"}}]}
I, [2025-07-01T21:21:07.869923 #64307]  INFO -- : response: Status 200
I, [2025-07-01T21:21:07.870027 #64307]  INFO -- : response: date: "Tue, 01 Jul 2025 21:21:07 GMT"
content-type: "application/json"
transfer-encoding: "chunked"
connection: "keep-alive"
access-control-allow-origin: "*"
vary: "Accept-Encoding"
server: "cloudflare"
cf-ray: "9588f37d7e87c932-IAD"
I, [2025-07-01T21:21:07.870061 #64307]  INFO -- : response: 
         
{"id":"gen-1751404866-7raGSzOBJFRZdmagdAhS","provider":"Google","model":"anthropic/claude-sonnet-4","object":"chat.completion","created":1751404866,"choices":[{"logprobs":null,"finish_reason":"stop","native_finish_reason":"stop","index":0,"message":{"role":"assistant","content":"I've successfully used the available tool. The tool has been called and executed as requested.","refusal":null,"reasoning":null}}],"usage":{"prompt_tokens":415,"completion_tokens":21,"total_tokens":436}}
"I've successfully used the available tool. The tool has been called and executed as requested."

Sorry... I'm likely just missing some key piece of documentation? I think I was expecting it to not have that final response message. Is the expectation that I should discard the final assistant message?

[joshforbes]

Upon further reflection, fair chance I'm misunderstanding this method. It does say and_respond. So maybe it just halts future possible tool calls, but does still respond based on the output of the current tool.

Given that, I assume there isn't a way to stop the tool call without an LLM response? Was hoping to have a way to essentially "pause" the chat for an indeterminate amount of time. Waiting minutes/hours/days/whatever for the user to approve/deny the suggested course of action. Upon approval, run some code and feed the output back into a rehydrated chat as the tool response.

[joshforbes]

Mmm actually something may be wrong? As soon as I add that stop_tool_calls_and_respond! it does away with all functions:

vs

[joshforbes]

Figured there was still a good chance that I was missing something critical, so I tried the example from readme:

But same result:

Am I doing something wrong?