GitHub Copilot - Sonnet - Rate Limit exceeded Error Due to Fair Usage

[GaryJurgens]

Which version of the app are you using?

Latest

Which API Provider are you using?

VS Code LM API

Which Model are you using?

Sonnet 3.5

What happened?

hi

I am Encountering , rate limit Errors, not due to the Frequency of Requests but due to Fair Usage Policy's, I have GitHub Enterprise, i am the only User, there are no enterprise or organization imposed limits.

There are 2 Issues at play,

1. Sonnet starts Hallucinating at anything above 500K tokens, per task

2. Between 500k and - 1 million Tokens, the fair usage policy kicks in on requests.

I am Using Lasted Release as of today, using Memory bank by GreatScottyMac/roo-code-memory-bank

Point Form, rigorous descriptions and requirements, for design and code. Emphasized no Assumptions, or hallucinating, check and stop if detected.
Checked RooCode understanding of solution from MeomoryBank and made it compare to its understanding by reviewing project structure, it has a good understanding.

Solution
Automatically Create new Tasks or possibly subtasks if task will exceed or starts to exceed 500k tokens.

I have Tried Direct Model Rules,
Having Roo Create an MCP server and associated wire up.
"Power Steering" does little to correct this issue, or realign the model and keep it in check.

It keeps going on and on in the same task, causing Hallucinations and self reinforcing loops. completely decimating the codebase. what built successfully , is totally ruined by new task.

I believe this is a bug, that needs to be addressed. Because it makes both my hours of manual code , and any previous RooCode completely decimated. i cannot restrict it enough.

C# and .Net Core

I will gladly assist in any way.

Steps to reproduce

Relevant API REQUEST output

Additional context

No response

[bitnom]

Have you done any code debugging on the issue or just user testing? I just started diving into vscode copilot integrations for a different project so I assume I'll run into this issue next. I'm happy to contribute a solution if I can find one.

I'll need to investigate how vscode LM API and roocode handle context management beyond the model's context length under the hood.

[GaryJurgens]

I am actively trying to look for a solution, perhaps an MCP server. I am running into the Same problem this Morning with the new Sonnet 3.7, its the Acceptable use Policy and not API call frequency. Because I think the task becomes to large.