openclaw - 💡(How to fix) Fix [Feature]: Allow selection of multiple models and one default during onboarding for reducing cost of openclaw [1 participants]

Summary

Enable users to configure multiple LLMs during onboarding instead of being limited to a single model.

Problem to solve

Currently, OpenClaw onboarding only allows the selection of one default model. This creates inefficiencies in real-world usage where different tasks require different levels of model capability.

Users often need to balance cost, latency, and performance:

Simple tasks (e.g., formatting, short responses) do not require powerful or expensive models Complex tasks (e.g., reasoning, long-form generation) benefit from more advanced models

With only one default model, users are forced to either:

Always use a high-cost model (wasting resources), or Manually switch configurations outside the onboarding flow (adding friction and inefficiency)

This limitation prevents optimized usage and increases operational costs unnecessarily.

Proposed solution

Allow users to configure multiple models during onboarding and one default Other model to enable > and then select

Suggested improvements:

Multi-model setup during onboarding Users can register multiple LLM endpoints (local or remote) Model selection from the desktop/UI Ability to switch models depending on task complexity easily

Future enhancement (optional) Automatic model routing based on task complexity Heuristics or lightweight classification to choose the most cost-efficient model

This approach would allow users to optimize both performance and cost without friction.

Alternatives considered

Single default model + manual reconfiguration This is the current approach and creates unnecessary friction and inefficiency. External orchestration (user-managed switching) Possible but requires additional tooling and defeats the purpose of a streamlined onboarding experience. Forcing users to always use high-performance models Simplifies UX but significantly increases costs and reduces flexibility.

Impact

Affected users/systems/channels

All users working with multiple LLMs Power users optimizing cost/performance Teams running mixed workloads

Severity

Moderate to high (does not block usage but creates persistent inefficiency)

Frequency

Very frequent (affects nearly every session involving varied task complexity)

Consequences

Increased operational costs Inefficient resource usage Extra manual steps to switch models Reduced usability for advanced workflows

Evidence/examples

Common real-world workflow:

Use a low-cost model for: Text formatting Simple Q&A Use a high-performance model for: Complex reasoning Long or structured outputs

Other platforms and orchestration tools already support multi-model strategies to optimize cost vs performance, indicating clear demand for this capability.

Additional information

This feature aligns with cost-efficiency goals and scalable usage patterns. It would also lay the groundwork for future intelligent routing systems that automatically select the most appropriate model based on task complexity.