dsh-llm-commandcode

Configuration

Field	Default	Meaning
apiKeyEnv	COMMAND_CODE_API_KEY	Credential reference resolved per request
baseURL	https://api.commandcode.ai	API root; /alpha/generate is appended
workingDir	process working directory	Reported to the endpoint as request context
cliVersion	installed command-code version, else 1.53.0	Sent as x-command-code-version, which the endpoint gates on
maxTokens	64000	Default output cap; a catalog model's own cap and an explicit request value win
defaultContextWindow	1000000	Capacity for a model the catalog does not size
defaultReasoningEffort	high	Effort used while a session has picked none
streamIdleTimeoutMs	300000	Maximum silence between two stream events
models	the five DeepSeek models	Static advisory catalog, used when discovery is off or unreachable
modelCatalog	auto	auto refreshes from the endpoint (cached an hour); static never asks
zeroDataRetention	false	Send x-cmd-zdr: 1, asking the gateway for zero retention
requestImagePixelBudget	640000	Route default pixel budget per request image; a catalog model's own budget wins
requestImageMaxBytes	1048576	Route default encoded-byte target per request image; a catalog model's own target wins
maxRequestImageBytes	20971520	Bound on accumulated base64 image payload in one request
maxImagesPerRequest	600	Maximum represented images in one request
imageOffloadByteQuantum	10485760	Raw-byte removal step once the payload bound is exceeded
imageOffloadCountQuantum	20	Image-count removal step once the count bound is exceeded
retryPolicy	normal, five retries	Provider-owned retry policy

A per-model override goes on the catalog entry itself:

- id: llm-commandcode
  name: 'dsh-llm-commandcode'
  config:
    models:
      - id: deepseek/deepseek-v4.1-flash
        inputModalities: [text, image]
        imagePixelBudget: 1048576 # a larger per-image budget than the route default
        imageMaxBytes: 2097152