Skip to content

Document local-dev models setup matching production #596

Description

@heskew

The last unwritten piece of HarperFast/harper#510's documentation acceptance ("Documented: writing a custom backend, configuring fallback, local-dev setup matching production"):

Scope: a page under reference/models/ covering:

  • The portability model — application code addresses models by logical name; the local ↔ production swap is configuration-only.
  • A local configuration with the ollama backend (no credentials) and its matching production configuration (hosted providers + ${VAR} credential indirection).
  • Parity caveats: the ollama backend does not advertise tools (workaround: openai backend + baseUrl against an OpenAI-compatible local server such as vLLM); embedding vectors are not comparable across different models — matters when @embed data or HNSW indexes move between environments.
  • Per-environment configuration: per-instance config files; HARPER_CONFIG merge for containerized deployments.
  • Offline / CI stubs via models.registerBackend / models.defineBackend (a registered backend shadows a same-name config entry).

Part of #548 (v5.1 coverage umbrella). Docs deliverable for HarperFast/harper#510.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions