dsh-notemd

Profile configuration

The bundle patch defaults stateful providers to process.cwd(). A deployment profile must replace the whole config object for every row it overrides; DSH patches do not deep-merge rows. Keep the complete field set:

- id: notemd-vault
  config:
    workspaceRoot: !!js process.env.NOTEMD_WORKSPACE_ROOT

- id: notemd-jobs
  config:
    workspaceRoot: !!js process.env.NOTEMD_WORKSPACE_ROOT
    concurrency: 2

- id: notemd-workspace-changes
  config:
    scanIntervalMs: 5000

- id: notemd-approval
  config:
    workspaceRoot: !!js process.env.NOTEMD_WORKSPACE_ROOT
    approvalTtlMs: 300000

- id: notemd-research
  config:
    workspaceRoot: !!js process.env.NOTEMD_WORKSPACE_ROOT

- id: notemd-artifacts
  config:
    workspaceRoot: !!js process.env.NOTEMD_WORKSPACE_ROOT

- id: notemd-llm
  config:
    provider: deepseek
    model: deepseek-chat
    maxTokens: 4096
    promptPolicyId: notemd.default.v1

The default notemd-llm provider injects DSH llm. Its closed route policy accepts only provider, model, maxTokens, and promptPolicyId. Endpoints, keys, headers, transport retries, and model discovery are rejected rather than ignored. Configure credentials, adapters, and provider selection in DSH; NoteMD never reads or persists them.

The explicit dsh-notemd/llm-openai-compatible-legacy entry is migration-only. It provides the former OpenAI-compatible diagnostic and model-discovery tools for deployments that cannot yet use DSH routing. Replace the default notemd-llm row when using it; never load both because both provide notemdTextTransformer.