[Bug]: AsyncStream' object has no attribute 'choices'

[kjekro]

Bug Description

Tracebacks in the log:

2026-04-27 18:48:49,619 - openviking.session.compressor_v2 - ERROR - Failed to extract memories with v2: 'AsyncStream' object has no attribute 'choices'
Traceback (most recent call last):
File "/home/kjekro/.local/share/pipx/venvs/openviking/lib/python3.13/site-packages/openviking/session/compressor_v2.py", line 196, in extract_long_term_memories
operations, tools_used = await orchestrator.run()
^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/kjekro/.local/share/pipx/venvs/openviking/lib/python3.13/site-packages/openviking/session/memory/extract_loop.py", line 202, in run
tool_calls, operations = await self._call_llm(messages)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/kjekro/.local/share/pipx/venvs/openviking/lib/python3.13/site-packages/openviking/session/memory/extract_loop.py", line 366, in _call_llm
response = await self.vlm.get_completion_async(
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
...<3 lines>...
)
^
File "/home/kjekro/.local/share/pipx/venvs/openviking/lib/python3.13/site-packages/openviking/telemetry/tracer.py", line 356, in async_wrapper
return await func(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/kjekro/.local/share/pipx/venvs/openviking/lib/python3.13/site-packages/openviking/models/vlm/backends/openai_vlm.py", line 404, in get_completion_async
return await retry_async(
^^^^^^^^^^^^^^^^^^
...<4 lines>...
)
^
File "/home/kjekro/.local/share/pipx/venvs/openviking/lib/python3.13/site-packages/openviking/utils/model_retry.py", line 129, in retry_async
return await func()
^^^^^^^^^^^^
File "/home/kjekro/.local/share/pipx/venvs/openviking/lib/python3.13/site-packages/openviking/models/vlm/backends/openai_vlm.py", line 398, in _call
return self._build_vlm_response(response, has_tools=True)
~~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/kjekro/.local/share/pipx/venvs/openviking/lib/python3.13/site-packages/openviking/models/vlm/backends/openai_vlm.py", line 179, in _build_vlm_response
choice = response.choices[0]
^^^^^^^^^^^^^^^^
AttributeError: 'AsyncStream' object has no attribute 'choices'

Steps to Reproduce

Install with pix in linux nas
openviking 0.3.13.dev0

OpenViking Doctor:

Config: PASS /home/kjekro/.openviking/ov.conf
Python: PASS 3.13.5 (>= 3.10 required)
Native Engine: PASS variant=x86_avx2
AGFS: PASS AGFS SDK 0.1.7
Embedding: PASS openai/Qwen3-Embedding-0.6B-Q8_0.gguf
VLM: PASS openai/Qwen2.5-VL-7B-Instruct-Q5_K_M.gguf
Ollama: PASS not configured
Disk: PASS 155.3 GB free in /home/kjekro/.openviking/data

All checks passed.

Expected Behavior

No traceback

Actual Behavior

Minimal Reproducible Example

.\llama-b8946-bin-win-cpu-x64\llama-server.exe `
  -m Qwen2.5-VL-7B-Instruct-Q5_K_M.gguf `
  --host 192.168.10.123 `
  -ngl 0 `
  -c 16384 `
  --port 8082 `
  --mmproj Qwen2.5-VL-7B-Instruct-Q5_K_M_mmproj-F16.gguf `
  --jinja

+

.\llama-b8946-bin-win-cpu-x64\llama-server.exe `
  -hf Qwen/Qwen3-Embedding-0.6B-GGUF `
  --embeddings `
  --host 192.168.10.123 `
  --port 8081 `
  -c 8192 `
  -b 512 `
  -ub 512 `
  --n-gpu-layers 0 `
  --pooling last

Error Logs

OpenViking Version

0.3.13.dev0

Python Version

3.13.5

Operating System

Linux

Model Backend

None

Additional Context

ov.log
ovconf.txt

[onthebed]

I've been looking into this — I want to trace the async response path and see where streamed chunks stop matching the expected choices shape. Happy to send a fix.