ChatWalaʻau v0.109.0
This release moves the five per-task "helper" models into the Model Offering Catalog as task model assignments, and removes the dedicated per-task model environment variables. This is a breaking change -- read the upgrade note below before upgrading. It continues the catalog-only model that 0.107.0 introduced for chat and 0.108.0 for image / embeddings.
Highlights
Task model assignments (v0.109.0)
- The helper models used to generate a chat's title, extract user memory, curate agent memory, summarize a Teams meeting, and convert a natural-language question into an ontology SPARQL query are now assigned in
model_offerings.jsoncvia a new top-levelrolesblock. Each role points at one of your existing chat offerings, so it inherits the correct provider, endpoint, and credentials automatically. - Assign them three ways: the in-app Model Settings screen (a new Task model assignments section), the CLI (
chatwalaau models role list | set | clear), or by hand inmodel_offerings.jsonc. - An unassigned task keeps the previous behavior: it follows the chat's own model, then the catalog default. A role that points at a deleted or non-chat offering falls back to that same default (and is flagged in the UI and a startup warning) rather than failing.
A silent mis-route is fixed (v0.109.0)
- Since the catalog became the exclusive router (0.107.0), a per-task model set to a bare deployment name that was not a catalog offering could be silently routed to Azure OpenAI -- wrong on an Anthropic-, OpenAI-, or gateway-only deployment. Because a role now references a real chat offering, task models route to the correct provider by construction.
Removed and no longer read (v0.109.0)
SESSION_TITLE_MODEL->roles.session_titleUSER_MEMORY_EXTRACTION_MODEL->roles.user_memory_extractionAGENT_MEMORY_CURATION_MODEL->roles.agent_memory_curationTEAMS_MEETING_SUMMARY_MODEL->roles.meeting_summaryONTOLOGY_NL_MODEL->roles.ontology_nl
Retained and unchanged (v0.109.0)
- Every non-model setting of these features is kept:
SESSION_TITLE_MODE, the user- and agent-memory toggles and caps, the Teams-meeting transcript / output settings, and the ontology enable / directory / limit settings. DEMO_MODEdeployments are unchanged (the catalog and roles are ignored under demo). A deployment that never set any of the five variables sees no behavior change.
Upgrade notes (BREAKING)
- Only if you set one of the five variables: move the value into a
rolesentry -- Model Settings -> Task model assignments,chatwalaau models role set <role> <offering-id>, or by hand inmodel_offerings.jsonc. Until you do, that task uses the chat's own model (safe, not an error). - Leftover legacy variables in your
.envare harmless -- they are ignored, and a non-failing startup advisory names the successor role.
Included in this release
This release includes implementation work from:
- v0.109.0 through v0.109.0
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:
0.109.0 - Previous release:
0.108.0