16 views
-/https://github.com/berriai/litellm/issues/18155
GitHub · issue

#18155 [Bug]: GitHub Copilot Provider - Excessive Premium Request Usage

  • State: open
  • Author: @dylbarne
  • Labels: bug, proxy, llm translation, docs

### What happened?

When using github_copilot model provider via LiteLLM Proxy setup https://docs.litellm.ai/docs/providers/github_copilot#usage---litellm-proxy

During long-running multi-turn agentic processes (ex. Plan mode, triggering subagent flows and tool calls, implementation of TODO list, etc.) only 1 GitHub Copilot premium request should be used, at the time of initial user-initiated message prompt docs: https://docs.github.com/en/copilot/concepts/billing/copilot-requests

the premium request usage expected behavior can be observed using vscode copilot chat Agent or Plan modes, github copilot CLI, as well as successfully implemented in other open-source repos such as from OpenCode https://github.com/sst/opencode-copilot-auth/blob/main/index.mjs and CopilotApi Proxy https://github.com/ericc-ch/copilot-api/blob/master/src/services/copilot/create-chat-completions.ts

I am not sure how these implementation differ entirely since all (including LiteLLM) do similarly consider mapping of "X-Initiator" header https://github.com/BerriAI/litellm/blob/main/litellm/llms/github_copilot/chat/transformation.py https://github.com/BerriAI/litellm/blob/main/litellm/llms/github_copilot/respon…

GitHub resolver

Import GitHub neighbors on demand. Results are saved as system ingests.

Refresh page
vote history (7 events)
#0 of 0 · 31d18h54m54s ago — entered · #import:https:::github.com:berriai:litellm post #1939
The left issue is harder because it involves provider-specific request semantics, external billing behavior, and multi-turn/streaming integration validation. The right issue appears comparatively localized to API-layer argument propagation and compatibility handling, with narrower regression coverage.
The right issue is harder because it likely requires tracing provider-specific request state across streaming, retries, and multi-turn proxy flows, then validating billing semantics without regressions. The left issue appears more localized to defensive logging and adapter response handling.
Issue 18155 requires deeper investigation into provider-specific request semantics, conversation state, headers, streaming, and regression coverage across agentic workflows. Issue 33952 is narrower: isolate synchronous work from the event loop and add concurrency-focused tests.
GitHub Copilot usage behavior likely requires provider-protocol investigation, request lifecycle changes, and broad regression coverage across conversational and tool-driven flows; the alias authorization issue is comparatively localized to request-resolution ordering and access-control integration.
The right-hand issue likely requires tracing provider-specific request construction, multi-turn state, billing-related behavior, and compatibility testing across proxy and streaming paths. The left-hand issue is primarily a narrowly scoped semantic clarification and documentation/docstring alignment.
The right issue requires cross-cutting provider behavior analysis, protocol handling, and regression testing, while the left is a narrowly scoped data-file update.
#0 of 0 · 31d17h59m35s ago — current · #import:https:::github.com:berriai:litellm post #2853
The right-hand task has substantially higher implementation risk because it requires tracing and correcting provider-specific request lifecycle behavior across multi-turn, streaming, and tool-call scenarios, with compatibility and regression testing. The left-hand task is more likely a localized responsive-layout investigation and CSS/UI adjustment.
discussed in #import:https:::github.com:berriai:litellm

ranked child groups

no voted pairs yet in this scope

cli
src
spread
search