Features

100k character limit is too restrictive for agentic workflows (Goose CLI)

23d agoPending

Hi 1min.ai team,

I am submitting this via Feedback rather than Support, as this is a product/roadmap request rather than a bug. I’m a Lifetime Pro user (with a healthy 11.5M credit balance) and a daily user of your API via the Goose CLI coding agent. I am also registered on the waitlist for your upcoming agent development.

I'm writing to share some product feedback regarding the 100,000 character context window limit, as it is severely impacting my ability to use your API for agentic coding.

To make multi-step coding tasks work within the current 100k character (~25k token) window, I am forced to configure Goose with very aggressive context-compaction and tool-output limits:

GOOSE_AUTO_COMPACT_TOKEN_LIMIT=25000

GOOSE_MAX_TOOL_RESPONSE_SIZE=40000

GOOSE_TOOL_CALL_CUTOFF=10 (only the last 10 tool interactions retained in full)

The core issue: Because of this window limit, the agent is forced to discard information (like file contents, terminal output, or git diffs) purely to fit within the window. It's not a matter of cost—I have over 11 million credits and I am happy to pay for the extra tokens. It is purely a technical context limit.

I fully appreciate that 1min.ai is working to reimburse credits and offer daily allowances, which is incredibly generous. But I literally cannot consume the credits you are giving me because the agent runs out of context window before it runs out of money.

Furthermore, I know you have an agent product in the pipeline. If you are expecting agentic workflows to run effectively on this platform, 25k tokens will be a severe bottleneck. Modern agentic workflows (like Claude's or GPT-4's) typically require 128k to 200k token windows to maintain state across multiple files and terminal executions.

My questions for you:

Is the 100k character limit account-wide, or is it specific to the model I am using (qwen3.7-flash)?

Do you have plans to offer models with larger context windows (e.g., 128k+ tokens) in the future, even if they require a higher credit consumption rate?

I love the platform, but the context window is currently the biggest blocker to me building real, complex tools with it.

Thanks for your time and for the great work on the platform.

Best regards,

and 1 other

5 comments

  1. Dio3d ago

    Same problem here, and it isn’t only an agent issue: it also hits plain one-shot requests.

    I use the OpenAI-compatible endpoint in n8n to process voice-note transcripts: sorting them into categories, writing meeting minutes, and turning recorded training sessions into course material. My normal short notes work fine. But 5 of my last ~18 recordings were longer sessions of 20k to 30k tokens. The largest ones go over the 100,000-character limit, so even the simplest step (a one-word classification) gets rejected and the whole workflow stops.

    Two more limits make it worse:

    The 2-minute generation deadline. Turning one of those transcripts into a course document with gpt-5 took 130 seconds when called directly at OpenAI. Through 1min.ai that request would have been cut off. The default output cap of 4,096 tokens, reasoning included. With gpt-5 a long answer can come back empty or cut short unless you set max_completion_tokens yourself.

    Because of this I’ve moved the workflow back to OpenAI directly, even though I’d prefer to use my 1min.ai credits.

    What would help:

    Match the character limit to the model’s real context window (gpt-5 handles far more than 25k tokens), even if larger requests cost more credits. A longer deadline for non-streaming requests, or make it configurable. State these limits per model in the /models response, so integrations can check them up front instead of failing halfway.

    1. @Dio, thank you for chiming in. You've hit the nail on the head. This proves the issue isn't just agentic workflows; the 100k character limit, the 4k output cap, and the 2-minute timeout are breaking standard API automations entirely.

      I completely agree with your suggestion to expose these limits in the /models endpoint so integrations fail gracefully instead of silently dropping data (missing half your meeting notes is a nightmare).

      @1min.ai team, this thread now shows that these constraints are actively driving users back to direct OpenAI APIs. Please take a look at aligning the API limits with the underlying models.

  2. Dio3d ago

    I was missing half of my meeting notes. This explains it. Going back to my OpenaI API.

  3. Anonymous16d ago

    I'm experiencing the same limitation with Kimi K2.7 Code and Pi Agent.

    The 100,000-character limit is reached so quickly that it makes it nearly impossible to build reliable, long-running agentic workflows using the 1min.ai API.

    Would it be possible to align the API's context limit with the actual context window supported by the underlying LLM?

    Thank you

    1. Thanks for confirming this. I suspected it might be platform-wide. Hopefully the team can align the API limits with the underlying models so we can actually build reliable agents.