mirror of
https://github.com/hanzoai/insights.git
synced 2026-10-30 16:15:31 +00:00
zeekay
afec777d0c
Pulse synthesis, heatmap grounding, AI-subscription spec generation and anomaly investigation all imported `ee.hogai.llm`. That package is not in this fork, so all four have been unimportable — dead on arrival, not degraded. They do not need porting. products/insights_ai/backend/model.py already holds the native client: a ChatOpenAI pointed at HANZO_API_URL with this deployment's IAM bearer, taking the same user/team/billable that the ee one took. It is the seam the assistant already uses, so this is four call sites joining the one way in rather than a fifth way appearing. MaxChatAnthropic goes with them. The gateway speaks the OpenAI dialect for every model in its catalogue — the assistant drives claude-haiku-4.5 through exactly this client — so one class covers both call styles and the second one was only ever an upstream detail about which vendor SDK to import. Leaves report_pipeline dead: it also wants ee.hogai.context and ee.hogai.tool_errors, which have no native counterpart yet. Fixing what is fixable rather than half-fixing all of it.