Openagents launcher

[Abz-Alharbi]

I am currently using OpenAgents Launcher and would like to build multiple agents connected to models through OpenRouter.

My goal is to create separate agents using different models such as Kimi, Qwen, DeepSeek, GLM, and others by using the OpenRouter API endpoint.

Current configuration:
• API Base URL: https://openrouter.ai/api/v1
• Provider type: OpenAI-compatible / supported addon inside OpenAgents Launcher

The issue I am facing is that model settings appear to be shared globally across agents.

For example:
1. I create Agent 1 and assign it to Qwen.
2. I create Agent 2 and assign it to DeepSeek.
3. After saving Agent 2, Agent 1 automatically changes from Qwen to DeepSeek as well.

This means I cannot keep different models assigned to different agents at the same time.

Could you please clarify the following:
1. Does OpenAgents Launcher currently support assigning different models to different agents independently?
2. If yes, what is the correct way to configure multiple agents with separate OpenRouter models?
3. If not, is this feature planned in a future update?
4. Is there a recommended workaround for using multiple OpenRouter models simultaneously inside OpenAgents Launcher?

My intended use case is to run multiple specialized agents, each powered by a different model.

Thank you for your time and support. I would appreciate any guidance you can provide.

[linear]

OADEV-56

[QuanCheng-QC]

Thanks for the detailed report — I took a look at the codebase and can confirm what you're seeing is expected given the current design.

1. Current behavior (confirmed)

OpenAgents Launcher does not support per-agent model configuration for agents of the same type.

Model/env settings are stored per agent type, not per agent instance. Specifically:

• Env files are saved at: ~/.openagents/env/<type>.env
• Launcher uses saveAgentEnv(type, env) when creating/updating agents
• The configuration UI is also bound to agentType, not individual agent instances

Because of this, multiple agents using the same provider/addon (e.g. OpenAI-compatible / OpenRouter) will share the same model configuration.
So updating Agent 2 will overwrite the shared config and affect Agent 1.

---

2. Is per-agent configuration supported?

• Launcher UI: ❌ Not supported
• Runtime (daemon): ✅ Partially supported

The runtime actually merges:

1. type-level env
2. agentCfg.env (per-agent override)

However, Launcher UI does not expose a way to edit per-agent env, so this capability is effectively hidden.

---

3. Workarounds

Depending on your use case, here are the available options:

Option A (recommended): Use SDK / YAML agents

Define agents explicitly with their own model config:

agent_id: qwen-agent
config:
  model_name: "qwen/your-model"
  provider: "openrouter"
  api_base: "https://openrouter.ai/api/v1"

agent_id: deepseek-agent
config:
  model_name: "deepseek/your-model"
  provider: "openrouter"
  api_base: "https://openrouter.ai/api/v1"

Avoid using model_name: auto.

---

Option B: Use different agent types

Since env is isolated per type, using different runtimes (e.g. openclaw, codex, nanoclaw) allows different models.

---

Option C (advanced): Manual per-agent env override

You can manually add env to specific agent entries in daemon.yaml, since the runtime supports merging agentCfg.env.

However:

• Not managed by Launcher UI
• May be overwritten or become inconsistent

---

4. Suggestion / potential improvement

Since the runtime already supports per-agent env overrides, this seems primarily a Launcher UI limitation.

A possible improvement would be:

• Keep type-level config as default
• Add optional per-agent override in the UI (writing to agentCfg.env)

This would enable multiple agents of the same type to use different models without breaking existing behavior.

---

Summary

• The issue description is accurate for Launcher
• Behavior is by design (type-level config), not a bug
• Per-agent configuration exists at runtime but is not exposed in UI

Hope this helps clarify