#Change Dreaming model
1 messages ยท Page 1 of 1 (latest)
/dreaming only toggles it on/off.
If you want a different LLM for dreaming itself, set it in config under the memory plugin, usually memory-core:
{
plugins: {
entries: {
"memory-core": {
config: {
dreaming: {
enabled: true,
execution: {
defaults: {
model: "openai/gpt-5.4",
},
},
phases: {
deep: {
execution: {
model: "anthropic/claude-opus-4-6",
},
},
},
},
},
},
},
},
}
Notes:
execution.defaults.model= default for all dreaming phasesphases.light|deep|rem.execution.model= per-phase override- restart after changing config:
openclaw gateway restart - this is JSON5, not YAML
Docs:
- https://docs.openclaw.ai/concepts/dreaming
- https://docs.openclaw.ai/reference/memory-config#dreaming
If you want, paste your current dreaming config and I can rewrite it for the model you want.
If this fixed it, please right-click the solution message โ Apps โ โ
Mark Solution, so Answer Overflow can index it ๐
How to change all phase dreaming and everything of dreaming to 1 model
Yep, set only the default dreaming model and remove any per-phase overrides.
{
plugins: {
entries: {
"memory-core": {
config: {
dreaming: {
enabled: true,
execution: {
defaults: {
model: "openai/gpt-5.4",
},
},
},
},
},
},
},
}
Important:
- use
execution.defaults.model - do not set:
phases.light.execution.modelphases.deep.execution.modelphases.rem.execution.model
If those phase-level model keys exist, they override the default.
Then restart:
openclaw gateway restart
If you paste your current dreaming block, I can clean it up into the exact final config for one model.
"memory-core": {
"config": {
"dreaming": {
"enabled": true
}
}
},
Add this inside your existing dreaming block:
"memory-core": {
"config": {
"dreaming": {
"enabled": true,
"execution": {
"defaults": {
"model": "openai/gpt-5.4"
}
}
}
}
}
Replace openai/gpt-5.4 with the model you want.
Since you currently have no phase overrides, this will make all dreaming phases use that one model.
Then restart the gateway:
openclaw gateway restart
If you want, I can also show the same snippet for a specific provider/model name you plan to use.