Automation runs always use K2.6 Agent, ignoring all model settings

Product Feedback
1

Product: Kimi desktop app (Windows) — Kimi Work / Blueprint Automations
Date: 2026-07-27

Summary

Every Agent Automation run (local_conversation / cron job) creates a brand-new conversation pinned to the factory-default model K2.6 Agent (k2d6-agent). There is no user-facing way to make automation runs use K3 Swarm (k3-agent-ultra) with Max thinking, even though my interactive chats in the same workspace use it.

What I verified

  1. Each run creates a NEW conversation. Runs never reuse a conversation, so per-conversation model settings can never apply to them.

  2. Run sessions always run k2d6-agent. The kernel session logs of every automation run show model k2d6-agent, while my interactive chats in the same workspace run k3-agent-swarm at the same time.

  3. The workspace/project default model is ignored. The project default for the automation’s workspace is set to k3-agent-ultra, yet runs still use k2d6-agent.

  4. The runtime “current” model is ignored. The daemon’s current model was k3-agent-swarm at the moment a run fired; the run still used k2d6-agent.

  5. Manually triggering an automation from a K3 Swarm chat does not inherit the model — the manual run also used k2d6-agent.

  6. Changing the model inside a run’s conversation in the UI opens a new chat instead of switching the existing conversation, so there is no UI path either.

  7. There is no “default model” option in Settings — neither for automations nor globally.

Expected behavior

Automation runs should honor the workspace/project default model (or expose an explicit “automation model” setting), so agent runs can use K3 Swarm + Max thinking like interactive chats in the same workspace.

Actual behavior

All automation runs — scheduled, condition-triggered, or manually triggered — always start with k2d6-agent, regardless of every user-visible model setting.

2

Same here; I haven’t figured out a way to get K3 to run as Work.