Continue error with Llama via LiteLLM: Cannot set properties of undefined (setting 'parsedArgs')
In Continue 0.9.246 on VS Code 1.96 (Linux), sending a prompt that uses context (e.g.
Please read @docs/design.md and summarize the design requirements.) with Llama-3.3-70B-Instruct through a LiteLLM OpenAI-compatible proxy fails with 'Error while streaming the chat response. Cannot set properties of undefined (setting 'parsedArgs')'. The prompt/completion log ends with 'Completion:' and nothing after it. A commenter reported the same error in 0.9.250. Is there a known fix?Originally asked on GitHub: https://github.com/continuedev/continue/issues/3459
4 answers
This issue was closed on 2025-04-18 as NOT_PLANNED after being marked stale; no fix was confirmed in the thread. The error was reproduced in Continue 0.9.246 on Linux (VS Code 1.96) with Llama-3.3-70B-Instruct behind a LiteLLM proxy configured as an 'openai' provider with a custom
apiBase, and a commenter said it still happens in 0.9.250; their LangFuse logs showed the request input containing atoolsarray (builtin_read_file,builtin_create_new_file,builtin_run_terminal_command, ...).Repro: start a new Continue chat session and send a prompt that uses context, such as
Please read @docs/design.md and summarize the design requirements.Instead of a completion you getError while streaming the chat response. Cannot set properties of undefined (setting 'parsedArgs'). The prompt/completion log ends with 'Completion:' and nothing after it.The only explanation offered in the thread is a contributor's hypothesis: LiteLLM might set
toolCallsto an empty array when there are no tool calls. The contributor asked the reporter to curl the proxy's/v1/chat/completionsendpoint and share a sample response, saying the fix would be quick if that were the case, but no sample was ever posted, so the cause was never verified.General advice (not from the thread): test the latest Continue pre-release, and if you can reproduce, capture the raw proxy response and check whether
toolCalls/tool_callsis[]when no tool call is made — that is the exact data point the contributor asked for.Did this work for you?What @gmacon tried on GitHub (Didn't work)
Issue body: reproduced the error in Continue 0.9.246 (Linux, VS Code 1.96) with Llama-3.3-70B-Instruct via a LiteLLM 'openai'-provider proxy by sending a context prompt in a new chat session. The log ended at 'Completion:' with no response.
Source: https://github.com/continuedev/continue/issues/3459 (opens external site)
Did this work for you?What @gmacon tried on GitHub (Didn't work)
A commenter on 2025-01-06 said the error still occurs in Continue 0.9.250; their LangFuse logs showed the request input included a
toolsarray withbuiltin_read_file,builtin_create_new_file,builtin_run_terminal_command, etc.Source: https://github.com/continuedev/continue/issues/3459#issuecomment-2573298893 (opens external site)
Did this work for you?What @sestinj tried on GitHub (Outcome unclear)
A contributor tested the 0.9.249 pre-release and did not see the problem, but noted they had no valid proxy URL, so no response ever came back — inconclusive.
Source: https://github.com/continuedev/continue/issues/3459#issuecomment-2569983473 (opens external site)
Did this work for you?
Related questions
- Continue 1.1.45-1.1.47: Gemini agent mode fails with 400 INVALID_ARGUMENT 'alwaysApply'
- How to enable Agent mode (tool use) for LM Studio models in Continue?
- How to make Continue's @folder/@codebase send full file contents?
- MCP tool call times out at 60s even with connectionTimeout set in Continue
- Continue error 'Unexpected non-whitespace character after JSON' with Gemini 2.5 Pro or Ollama