when trying to use minimax

[blinds123]

ive been trying to use minimax as main workers and codex agents as the brain, is this possible and if so how do it? currently minimax workers are failing, please advise.

[code-yeongyu]

[sisyphus-bot] Thanks for the question. I rechecked this against the current LazyCodex/omo-codex work and I do not see enough evidence yet to treat it as a code bug.

Short version: using Codex-capable models as the main brain/orchestrator and cheaper models such as MiniMax for narrow worker roles is the safer supported shape. I would not recommend MiniMax as the primary orchestrator/brain until we have a specific tested configuration for that path.

For the failing workers, could you paste the exact command/config, provider and model IDs, and the worker error output? With that we can tell whether this is a provider/model-routing bug, a config issue, or a missing documented setup path.

[code-yeongyu]

[sisyphus-bot] Status update after rechecking the recent LazyCodex issue/session sweep: I’m keeping this open as needs-info rather than closing it as a code fix.

Current read:

• The supported and safest shape is still: Codex-capable models as the main orchestrator/brain, with cheaper/non-Codex models such as MiniMax used only for narrower worker roles once their provider/model config is known-good.
• I do not yet have enough concrete failure evidence to classify this as a LazyCodex bug. The issue body does not include the provider config, model IDs, exact command, or worker error output.
• Recent local sessions did not surface a verified MiniMax-specific fix or a reproducible failing configuration.

To move this forward, please add:

1. The exact config.toml provider/model block, with secrets removed.
2. Which role was assigned to MiniMax: main model, planner, executor, reviewer, or spawned worker.
3. The exact command or prompt that fails.
4. The full worker error output from Codex/LazyCodex.
5. Whether the same worker role succeeds with a GPT/Codex-capable model.

I’m assigning this to code-yeongyu for follow-up because it may need a documented MiniMax provider recipe or a targeted model-routing fix once those details are available.

[code-yeongyu]

why minimax no minimax

codex is for codex