what is the most compatible llm to be used with opencode?

[aivagrin]

Question

hitting limit always for context length in new session when switching etc 137k for deepseek etc.

how to automatically trim / compress command for previous context input or how to work around this?

[github-actions]

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

• prompt is too long unrecoverable #4845: prompt is too long unrecoverable - Directly addresses context length limits with Anthropic models and mentions needing to compact
• AT_APICallError: prompt token count of 130389 exceeds the limit of 128000 #2464: AT_APICallError: prompt token count exceeds the limit - Similar token limit issue blocking workflow
• Context length issue / better context inspector #2841: Context length issue / better context inspector - Questions about hidden context usage and tracking
• Feature: generic /compact command, auto-compaction, and fork-aware conversations #4317: Feature: generic /compact command, auto-compaction, and fork-aware conversations - Proposes a comprehensive solution for automatic context trimming/compression
• Fetched documentation exeeds context window limit - Automatic truncation possible? #1212: Fetched documentation exceeds context window limit - Automatic truncation possible - Early request for automatic truncation of context
• Feature: Subagents should support auto compaction #3153: Feature: Subagents should support auto compaction - Related feature request for auto-compaction

These issues cover both the problem you're experiencing (hitting context limits with DeepSeek) and several proposed solutions (auto-compaction, context compression). Feel free to ignore if your specific case involves DeepSeek compatibility concerns not addressed by these.

[rekram1-node]

@aivagrin are you defining deepseek in your own opencode.json? there are a lot of things that could cause this

Here are some models we think you'll like:
https://opencode.ai/docs/models/#recommended-models

This ins't complete, there are a lot of options to choose from

[Cyrille37]

Hi
I'm joining this discussion, I hope I'm not disturbing anyone ;-)

About compatibility between AI coding tool :

I'm running llama.cpp on a RTX3060 12Go with "Qwen2.5-coder-7b-instruct-q8_0" model.
When asking to "write method's comments in @"

• cline well do the job (it does too much like git commit, I have to add rules...)
• opencode reply "Understood. bla bla" or it writes a new Unit test, but never add comment to methods

Is because of system prompt ? Other tips ?

Thanks & regards

[rekram1-node]

I would need to see the session to see what went wrong