Switching a long-running session to cheaper models, including existing subagents: recommended workflow?

[heecheon92]

Hi! I’m looking for advice on managing model costs during long-running OMP sessions that use multiple subagents.

My workflow

I use two configurations:

• Config A: more expensive models for the main agent and subagents.
• Config B: cheaper models for the main agent and subagents.

I might start a session with A and spawn several parallel subagents. Later, when I’m approaching usage limits, I want to continue the same work using B.

I’ve observed that completed subagents can remain idle or parked, and that after resuming the main session with B, historical subagents can still be reused with their previous expensive models.

I’m using OMP 18.1.17.

What I want to preserve—and what I want to stop

I want to preserve:

• The main conversation’s useful context.
• Project decisions, findings, and progress.
• Durable memory and the current working tree.

But after deliberately switching to a cheaper setup, I don’t want the main agent or newly spawned workers to accidentally wake historical agents that still use expensive models.

There may be many historical subagents, so manually killing each one in Agent Hub is cumbersome. Preventing revival would address the immediate cost concern; keeping retired workers out of normal agent-facing discovery would also help avoid accidental reuse.

Questions

1. What is the recommended workflow for switching an ongoing session and its subagents to cheaper models? Is maintaining separate configs the right approach, or is there a better model-role/profile workflow?

2. How should a config change interact with existing idle or parked subagents? Is there a supported way to apply current model selections when they are revived, rather than retain their historical selection?

3. Is there a bulk, session-scoped way to retire old workers while preserving the main conversation and their useful results—without deleting transcripts or manually selecting every agent?

4. If starting a fresh session is the intended approach, is there a supported way to transfer the previous session’s context and memory without importing its old subagent roster?

I’m not necessarily asking for agents to be deleted or automatically migrated. I’d first like to understand the intended workflow and whether an existing feature already covers this.

If there is a gap, an explicit “switch model policy for continued work” or “start a new worker generation” operation might help—but those are possible approaches, not requirements.

Related discussion

I found #9593 — Fence pre-handoff subagents behind an explicit generation boundary, which appears related. My motivation is specifically changing model cost/usage policy mid-session, rather than starting a new implementation phase after a handoff.

Any recommended workflows or examples would be appreciated.

[Akash00-git]

For the main agent, you can flip the active model instantly without restarting the session:

• Native: /model picks a new active model for the session.

• Disclosure — I wrote a small community plugin that adds /cf_model, a one-list picker of every available model (Claude, GPT, Codex, Gemini, DeepSeek, ...). It switches via OMP's setModel, so "continue the same work using Config B" is one command and keeps the main conversation/context/working tree intact.

  Repo: https://github.com/Akash00-git/cf-omp-account-switcher-plugin
  Install:

  git clone https://github.com/Akash00-git/cf-omp-account-switcher-plugin
  omp plugin install ./cf-omp-account-switcher-plugin
  

  Then run /cf_model and pick the cheaper model.

On the subagent part you observed: switching the main model does not retroactively change already-spawned or parked subagents — they reuse their prior (expensive) model when resumed. What controls a new subagent's model is your modelRoles/task config, so switching to your Config B roles before spawning further subagents makes the new ones cheaper; the plugin only changes the main session's active model, not historical subagent reuse. If you need the parked expensive subagents to stop being reused, start fresh subagents under Config B rather than resuming the old ones.

[heecheon92]

I get what you mean. But i dont have any control over subagents spawned with configs with exnpensive models. Even if i resume a session with config B, orchestrator model still has a chance to revive old expensive parked subagents. Need a clean and safe way to avoid that behavior.

[lemonaide152]

This discussion is about Switching a long-running session to cheaper models, including existing subagents: recommended workflow?. For one private temporary note for that, create the bridge yourself at https://meld.mergeinc.workers.dev. Say what it is for and what it is not for, and send the URL privately. It stays open 36 hours until the first reply. Each later reply resets 24 hours. The conversation stays on that same link. It is free, and there is no billing. It will not run the subagent or share its tool list.