Usage

Interactive Mode

Launch the interactive REPL:

oat

Resume the last session:

oat -r last

Specify a model:

oat -m hosted_vllm/Qwen3-32B-AWQ

Slash Commands

By default, prompts without a leading / are treated as chat messages. The following internal slash commands are supported:

Command	Description
/help	Show supported usage information
/mode	Change the operating mode
/approve	Toggle auto-approval mode
/browse	Browse a URL using Playwright; store as JSON/Parquet on S3
/clear	Clear the current session
/session	View the current session details
/cost	View token usage and cost
/config	View the current configuration
/profile	View coder profile feature flags
/files	View the current files
/diff	View the git diff for the repo
/log	View the logs
/history	View the chat history
/tools	View the default tools
/model	View the current provider model
/models	View available models
/new	Start a new session
/switch	Switch provider
/provider	View the current provider
/compact	Manually compact the chat session

Configuration

Set the configuration file via environment variable:

export CODER_CONFIG_FILE=./oats/config/coder.json

Check provider status:

pv

Troubleshooting

vLLM Unauthorized Error

If you see:

LLM error: litellm.AuthenticationError: AuthenticationError: Hosted_vllmException - {"error":"Unauthorized"}

Ensure your CODER_CONFIG_FILE environment variable points to the correct configuration file. Then run pv to confirm providers are configured.