[suggestion] save tokens by adding an ASK MODE

isn´t it crazy that a simple "hi" can cost 17.7K tokens ?
because of all the agents, tools/mcps....

what if everything already is in the context and i want to talk with the Ai about it.
i don´t need tools, agents, and all that stuff.

but sending a simple "hi" costs thousands of tokens, that´s crazy.

we currently have "PLAN MODE" and "BUILD MODE"
why not an "ASK MODE" where all tools, agents, mcps are ignored ?

we don´t need to send all this overhead with every single prompt, or am i missing something here ???

I second this, being able to just chat with your LLM is a most so that you don't have to go to chatgpt to ask to then go back a build stuff

Try this:

{
  "agent": {
    "ask": {
      "mode": "primary",
      "tools": {
        "*": false
      },
      "prompt": " "
    }
  }
}

If you don't have AGENTS.md, the only part of the system prompt that remains is

Here is some useful information about the environment you are running in:
<env>
  Working directory: ...
  Is directory a git repo: ...
  Platform: ...
  Today's date: ...
</env>
<project>
  
</project>

you can configure this yourself - we'll consider making it native but needs more thought

Try this:

{
"agent": {
"ask": {
"mode": "primary",
"tools": {
"*": false
},
"prompt": " "
}
}
}
If you don't have AGENTS.md, the only part of the system prompt that remains is
Here is some useful information about the environment you are running in:
<env>
  Working directory: ...
  Is directory a git repo: ...
  Platform: ...
  Today's date: ...
</env>
<project>
  
</project>

Thanks for the workaround! But there’s also a predefined prompt for an ‘empty’ agent. I just want to chat with the agent the same way I do on the ChatGPT website. no opencode prompt, no context etc.

Just adding some feedback from my workflow. I tend to find success with designing a Plan in Ask mode.

I have a conversation with the LLM about my goals and make decisions about those goals in the Ask mode. Then Compress that conversation into a Plan via Plan mode. Then when executing the plan in Build mode I will toggle to Ask mode if there isn't something quite right with the implementation or there is a bug. Ask is almost like a 2 Party thinking mode to make design decisions and take direction.

Now Ask mode needs the ability to scrape through the code base and then compress that information back into the conversation. So what seems to work well is a Sub-Agent of sorts that goes out and does the search work then returns information back into the main session's context.

Please for the love of god make ask mode native... and set it as the default mode. A simple git push to HF -> plan and build made this chonky thing happen! (I'm on the paid plan btw not that it matters)

One more vote to make it native, I believe that even that the terminal focus on hard users, we need to think that some people are hard users trying to save time (the payload of creating your own agent/skill) and tokens.

When coding for fun, I have 10 directories, I don't create an .agent folder for each one.

Hi ! Same requets here : a default "ask" agent would be very cool regarding my daily usage... thanks

Ask Agent should not be a blank prompt, but rather an Agent specifically designed for explaining code, with prompts like "Explain the code, do not edit", etc.

What you probably mean is Raw Agent.

But it's strange that opencode doesn't have Ask Agent built-in—after all, the need to use LLMs to assist in reading code is very common.

I find the built-in "explore" agent somewhat suit this usage. Perhaps make that agent "all" instead of "subagent"?

To stay organized issues are automatically closed after 60 days of no activity. If the issue is still relevant please open a new one.

Navigation Menu

Repository navigation

[suggestion] save tokens by adding an ASK MODE #1573

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

[suggestion] save tokens by adding an ASK MODE #1573

Description

Activity

davejfranco commented on Aug 4, 2025

anuramat commented on Sep 8, 2025

thdxr commented on Sep 17, 2025

shihaonan369 commented on Dec 27, 2025

cmcdowell03 commented on Jan 29, 2026

bitanath commented on Jan 31, 2026

fbzz commented on Mar 3, 2026

kbulabs commented on Apr 11, 2026

jmjoy commented on May 5, 2026

ylxdzsw commented on May 11, 2026

github-actions commented on Jul 10, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions