Agent seems unable to do anything using together.ai

This is my first attempt to get this software working. I am running Debian 12. I also use https://github.com/sigoden/aichat/ without problems using the same provider.

Using together.ai I can choose Qwen3 Coder, Kimi K2 etc okay. I can speak to the AI, but when I ask it to do anything, it says it will do it, but does nothing. For example:

I'll analyze the playbook and provide concise improvement suggestions. Let me read the file first:

cat install-playbook.yml

Then nothing happens. It is the same with the build and plan agent. I can continue talking to it and get responses like No problem, I'm working on analyzing the playbook. Let me read it now:, but nothing happens.

Not sure how to debug this? Should there be some pop-up to say if it is working in the background? Thanks!

$ opencode --version
0.5.13

This issue might be a duplicate of existing issues. Please check:

Local Ollama tool calling either not calling or failing outright #1034: Local Ollama tool calling either not calling or failing outright - Similar symptoms where agent says it will execute tools but nothing happens
Tool Calling Issues with Open Source Models in OpenCode #234: Tool Calling Issues with Open Source Models in OpenCode - Comprehensive issue about tool calling problems with open source models, including Qwen models
qwen3-coder-30B-A3B is not able to call any tool #1809: qwen3-coder-30B-A3B is not able to call any tool - Specific issue with Qwen3 Coder models not executing tools, getting type validation errors
Kimi K2 "AI_TypeValidationError" when using tools #914: Kimi K2 'AI_TypeValidationError' when using tools - Tool validation errors with Kimi K2 model
OpenRouter/ Kimi K2 just stops working #1329: OpenRouter/ Kimi K2 just stops working - Similar behavior where agent stops executing after saying it will do something
bug: DigitalOcean Interference fails to use tool-calls (openai-compatible provider) #800: DigitalOcean Inference fails to use tool-calls (openai-compatible provider) - Tool calling failures with OpenAI-compatible providers

The common pattern seems to be tool calling issues with non-Anthropic providers, particularly affecting Together.ai, OpenRouter, and other OpenAI-compatible providers when using open source models like Qwen and Kimi.

Feel free to ignore if none of these address your specific case.

what llms are you using

I tried Qwen3 Coder and Kimi K2 when I made the issue.I just tried Deepseek v3, same result, and Deepseek R1 and got:

Since the user wants a list of suggestions without changing anything, I will read the file and then provide
suggestions.  Let me analyze the playbook. I'll read the main playbook file:

cat install-playbook.yml

Once I have the content, I'll provide specific suggestions to simplify
it.<｜tool▁calls▁begin｜><｜tool▁call▁begin｜>function<｜tool▁sep｜>readFile

{"path": "install-playbook.yml"}
<｜tool▁call▁end｜><｜tool▁calls▁end｜>
Build deepseek-ai/DeepSeek-R1 (02:22 PM)

and again nothing happens. I am switching LLMs in the same session.

I started a new session with DeepSeek R1 in the same project directory, try to do the same thing and it is still not working, I get:

Let's read the file and then provide an analysis.

We'll use the ReadFile tool to read the entire file.

Since the file is at the root, the path is: install-playbook.yml  I'll analyze the playbook. Let me first examine
the main playbook file.<｜tool▁calls▁begin｜><｜tool▁call▁begin｜>function<｜tool▁sep｜>ReadFile

{"path": "install-playbook.yml"}
<｜tool▁call▁end｜><｜tool▁calls▁end｜>
Build deepseek-ai/DeepSeek-R1 (02:27 PM)

The cursor is flashing and I can still input questions.

Here are the models from together from models.dev https://github.com/sst/models.dev/tree/dev/providers/togetherai

I think this is an issue on their end, unless you are using some proxy, they aren't parsing out tool calls in a compatible manner it seems

@rekram1-node

@rekram1-node That is not good news for together.ai, I am not using any proxy, did you confirm they aren't parsing out tool calls in a compatible manner? I can contact their support.

Can you recommend another company that hosts multiple up-to-date open source models on their own servers, claims not to use your data, and is tested to work with opencode?

Azure seems pretty good, could also use bedrock, google vertex

and google copilot too

Of those I have only used Azure & google copilot, hard to keep up but I have:

openai
groq
cerebras
anthropic
azure
github copilot
google (not vertex)
openrouter

All of these work great for me, not all fit your specific criteria

[automated] Closing due to 90+ days of inactivity. Feel free to reopen if you still need this!

Navigation Menu

Repository navigation

Agent seems unable to do anything using together.ai #2169

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Agent seems unable to do anything using together.ai #2169

Description

Activity

github-actions commented on Aug 22, 2025

rekram1-node commented on Aug 22, 2025

David-Else commented on Aug 22, 2025

rekram1-node commented on Aug 22, 2025

David-Else commented on Aug 22, 2025

rekram1-node commented on Aug 22, 2025

rekram1-node commented on Aug 22, 2025

rekram1-node commented on Dec 28, 2025

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions