ChatWalaʻau v0.103.0
This release adds a new feature to ChatWalaʻau: the Model Offering Catalog -- an optional, file-based way to manage every model an instance serves.
Highlights
Model Offering Catalog (v0.103.0)
- Manage every model from a single JSONC file (
model_offerings.jsonc, setMODEL_OFFERINGS_FILE). Each "offering" declares one model: its provider, the real model / deployment name, an optional endpoint / base URL / hosting, a credential reference, its purpose (chat, embeddings, or image), an optional generation-option family, a context window, and an optional default flag. - Gateways made simple: a single endpoint that fronts several model families becomes several offerings sharing an endpoint -- no new provider to wire up.
- Mixed hostings in one instance: direct Claude and Foundry-hosted Claude can coexist (per-offering hosting), which the single global hosting switch could not express.
- One required chat model, everything else optional: at least one chat offering is required; a single embeddings offering feeds document search and a single image offering feeds image generation, in place of their dedicated variables.
- Secrets stay out of the file: offerings reference an environment variable by name (or use the shared Entra ID credential lanes); endpoints support
${VAR}substitution. A malformed catalog fails startup fast with a message naming the offending offering. - The startup log states the active source (
Model routing source: Model Offering Catalog (...)vs the legacy.envvariables), so it is obvious where models came from. - Fully backward compatible: with no catalog file (the default), the existing
AZURE_OPENAI_MODELS/ANTHROPIC_MODELS/OPENAI_MODELS/FOUNDRY_MODELSvariables keep working unchanged, and demo mode always uses them. When a catalog file is present it becomes the single source of truth and those variables are ignored for routing (a startup warning lists them). - Deprecation notice: the four
*_MODELSvariables and the globalANTHROPIC_HOSTINGswitch are deprecated in favor of the catalog and will be removed in a future release; migrate tomodel_offerings.jsoncwhen convenient. The endpoint and credential variables they reference are not deprecated.
Included in this release
This release includes implementation work from:
- v0.103.0 through v0.103.0
Notes
- Internal design management content and architecture control artifacts are intentionally omitted from this release note.
Installation
See the repository README for setup and usage instructions.
Version
- Release version:
0.103.0 - Previous release:
0.102.2