Troubleshooting¶
Command Not Found¶
If anna is not available after installation, check the tool directory used by
uv or pipx and ensure it is on PATH.
uv tool list
pipx list
Configuration Fails¶
Run:
anna doctor --workspace anna-workspace
anna config show --workspace anna-workspace
anna config validate --workspace anna-workspace
Check that settings.yaml exists and that secrets are present in .env.
Model Calls Fail¶
Verify the base URL, model name, and API key. The endpoint must be compatible with OpenAI chat completions.
anna test model --workspace anna-workspace
Memory Search Fails¶
Check embedding configuration first:
anna test embedding --workspace anna-workspace
anna memory stats --workspace anna-workspace
If the embedding endpoint changed, reset and rebuild the memory index.
vLLM Does Not Start¶
Use a longer wait timeout for large models:
anna models deploy --target complaint --backend vllm --workspace anna-workspace \
--gpu 0 --gpu-memory-utilization 0.85 --wait-timeout 900
If the CLI reports missing CUDA toolkit, confirm that nvcc is visible or pass
--cuda-home. On module-based clusters, AnnaAgent attempts to discover and
load CUDA modules for the child process automatically.
If FlashInfer JIT fails because ninja is missing, use a workspace deploy
environment. AnnaAgent will install ninja into that environment when needed.
Stale Logs Are Confusing¶
Each deploy resets the current target service log before starting vLLM. If you are reading logs manually, make sure you are looking at the target log under the current workspace, not an older workspace or previous run directory.
PyPI Install Fails¶
If installing from PyPI fails, verify the package name and Python version:
python --version
pip install anna-agent
AnnaAgent requires Python 3.10 or newer.