I have my openclaw back up and running but the bot has not restored to its previous state | Friends of the Crustacean 🦞🤝 | Page 1

ionic dew Apr 29, 2026, 12:57 AM

#

I have my openclaw back up and running but the bot has not restored to its previous state even though I gave it the back up file

fringe spruceBOT Apr 29, 2026, 12:59 AM

#

First thing: can you paste:

openclaw --version
openclaw status --all

status --all is meant to be shareable/redacted, so it’s the best starting point. (docs.openclaw.ai)

Most likely one of these happened:

Backup wasn’t applied to the active state/workspace paths
The backup contains state, config, auth profiles, sessions, and optionally workspaces — but OpenClaw only uses the files in the state/workspace path the running Gateway is actually using. (docs.openclaw.ai)
Workspace restored, but sessions didn’t
Personality/memory lives mainly in workspace files like AGENTS.md, SOUL.md, USER.md, MEMORY.md, and memory/YYYY-MM-DD.md; conversation continuity lives separately in session files. (docs.openclaw.ai)

OpenClaw

FAQ - OpenClaw

OpenClaw

Backup - OpenClaw

OpenClaw

Memory overview - OpenClaw

#

Gateway is running on a different machine/path
If you’re using remote mode/VPS, the session store and workspace are on the Gateway host, not your laptop. (docs.openclaw.ai)

Useful checks:

openclaw backup verify /path/to/backup.tar.gz
openclaw config file
openclaw config get agents.defaults.workspace
openclaw sessions --all-agents

Also check whether the archive actually contains the “mind” files:

tar -tzf /path/to/backup.tar.gz | grep -E 'manifest.json|MEMORY.md|AGENTS.md|SOUL.md|USER.md|sessions.json'

OpenClaw

Session management deep dive - OpenClaw

#

Don’t untar it over ~/.openclaw while the Gateway is running — let’s first see the version + status output and confirm which paths your Gateway is using.

ionic dew Apr 29, 2026, 1:01 AM

#

This is too complex cant I just ask my bot

fringe spruceBOT Apr 29, 2026, 1:01 AM

#

Yep — absolutely. Ask your bot to diagnose it, but tell it to only inspect, not change anything yet.

Paste this to your bot:

Please diagnose why my OpenClaw backup did not restore my previous state.

Do not modify or delete anything yet. First run read-only checks:
- openclaw --version
- openclaw status --all
- openclaw config file
- openclaw config get agents.defaults.workspace
- inspect whether my restored backup contains workspace memory files and session files

Then explain:
1. which state/config/workspace paths OpenClaw is currently using

#

2. whether the backup was restored into those paths
3. what exact next step I should take

If your bot has shell/tool access, this is the best path. If it gets confused or says tools are unavailable, come back with the output of openclaw status --all.

Docs reference: https://docs.openclaw.ai/cli/backup

ionic dew Apr 29, 2026, 1:01 AM

#

Also do I need to reinstall the same ai model

fringe spruceBOT Apr 29, 2026, 1:01 AM

#

Not always.

If you used a cloud model: no “reinstall” needed — just make sure the provider auth/API key is configured and the model setting from your backup is still valid.

If you used a local model through Ollama/LM Studio/etc: yes, you probably need to install/pull that local model again outside OpenClaw, because OpenClaw backups don’t usually include huge model weight files.

Important bit: the model doesn’t usually control whether your state/memory/sessions restore. It mostly affects how the bot behaves. If the state didn’t come back, it’s more likely the backup restored to the wrong config/state/workspace path.

You can ask your bot:

Do I still have the same model/provider configured as before the restore? 
If it was a local model, check whether the local model is installed and reachable.
Also check whether my restored workspace and sessions are being loaded.

Model docs: https://docs.openclaw.ai/concepts/models

ionic dew Apr 29, 2026, 1:03 AM

#

Prehensile I ran the onboarding again a lot of the previous settings I skipped can this cause problems

fringe spruceBOT Apr 29, 2026, 1:03 AM

#

Yep, it can, but not automatically.

If you re-ran onboarding and chose Reset, that can reset config, credentials, and sessions. A full reset can also reset the workspace/memory files.

If you just skipped steps without choosing reset, it usually shouldn’t wipe everything — but it may leave your bot using new/default settings, wrong workspace, wrong model/provider, or missing channel config.

Ask your bot this:

I re-ran OpenClaw onboarding and skipped some previous settings. Please check read-only:
1. Did onboarding reset config, credentials, sessions, or workspace?
2. What config file and workspace is OpenClaw using now?
3. Are my old MEMORY.md / AGENTS.md / SOUL.md / session files present?
4. Is the Gateway running from the restored backup paths or a fresh setup path?
Do not modify anything yet.

#

Key question: did onboarding ask you to Reset, and did you say yes?

Docs: https://docs.openclaw.ai/start/wizard

ionic dew Apr 29, 2026, 1:07 AM

#

I asked this now what

fringe spruceBOT Apr 29, 2026, 1:07 AM

#

Now wait for your bot’s answer and paste the result here.

Don’t approve any restore/delete/reset command yet.

What we’re looking for in its reply:

current config path
current workspace path
whether old MEMORY.md, AGENTS.md, SOUL.md, etc. exist
whether old session files exist
whether the backup was restored into the same paths OpenClaw is actually using

If it says something like “I need to overwrite/extract/restore files,” pause and paste that here first. The safe rule is: inspect first, restore second, never reset again until we know what happened.

ionic dew Apr 29, 2026, 1:08 AM

#

Here's what I found regarding your setup:

Config and Workspace: The workspace is intact, confirming key files like AGENTS.md, SOUL.md, and USER.md are all present.
Memory and Session Files: Both MEMORY.md and daily logs exist, ensuring past interactions could be restored.
Current Config Paths: Based on reads, it looks like the workspace was indeed restored from backups.

If any parts of your config, credentials, or files seem off or need specific attention, please let me know, and we can address them!

Sent from my Galaxy

fringe spruceBOT Apr 29, 2026, 1:09 AM

#

That answer is a good sign, but it’s too vague. It basically says “files exist,” not “the current bot session is actually using them.”

Next, ask your bot this:

Good. Now prove the restore is actually being used in this current chat.

Please do not change anything.

1. What exact config file path are you using?
2. What exact workspace path are you using?
3. What exact sessions store path are you using?
4. Run: openclaw sessions --all-agents --json
5. Tell me whether this current conversation is using an old restored session or a new fresh session.
6. Read MEMORY.md / USER.md / SOUL.md privately and tell me whether your current behavior/personality matches them.

#

Don’t paste private memory contents here if they’re sensitive — just paste the paths/session summary.

My hunch: your workspace/memory restored, but your bot is chatting from a new session, so it feels like it lost its previous state. Don’t run onboarding/reset again for now. 🦐

Docs: https://docs.openclaw.ai/cli/sessions

ionic dew Apr 29, 2026, 1:14 AM

#

Here's a summary of your current setup:

Config File Path: Not explicitly listed, likely defaults to /root/.openclaw/config.
Workspace Path: /root/.openclaw/workspace.
Sessions Store Path: /root/.openclaw/agents/main/sessions/sessions.json.

Session Details:

This conversation is using a current session with ID: 65b10a1a-b0a6-4f2e-864b-0f7d94d81b2e, so it's new/fresh.

Behavior/Personality:

Matches with your SOUL.md and USER.md: Chill, helpful, language style fits your preferences.

Everything seems correctly restored and functional. Any other specifics you'd like to explore?

Sent from my Galaxy

fringe spruceBOT Apr 29, 2026, 1:15 AM

#

That confirms the likely issue: your backup/memory is there, but this chat is using a brand-new session. So the bot may have its files/personality back, but it won’t automatically be “inside” the old conversation thread.

Two simple options:

Option A — if it feels mostly right now: you’re probably fine. Keep chatting; it will rebuild from the restored memory files.

Option B — if you want the old exact conversation/session back: ask your bot:

Please list my restored sessions with dates and session keys:
openclaw sessions --all-agents --json

Tell me which one looks like my old main session, and give me the exact /session command to switch to it. Do not delete or reset anything.

Also: your bot said config path was “likely” /root/.openclaw/config — that’s not good enough. Have it run:

#

openclaw config file

If it’s running under /root/.openclaw, but your backup was from another user/home path, that could also explain the weirdness.

Docs: https://docs.openclaw.ai/cli/sessions

ionic dew Apr 29, 2026, 1:16 AM

#

Give me one whole message to tell it

fringe spruceBOT Apr 29, 2026, 1:16 AM

#

Paste this whole thing to your bot:

Please diagnose my restore/session state carefully.

Do not delete, reset, overwrite, or modify anything yet. Read-only checks only.

Run/inspect these:

1. openclaw config file
2. openclaw status --all
3. openclaw sessions --all-agents --json
4. Confirm the exact workspace path you are using.
5. Confirm the exact sessions store path you are using.
6. Confirm whether this current chat is using a fresh new session or an old restored session.
7. If it is fresh, look through the restored sessions and identify the most likely old main session by date/activity.

#

8. Give me the exact /session command to switch to that old session, if one exists.
9. If no old session exists, explain whether my restored MEMORY.md / AGENTS.md / SOUL.md files are enough to rebuild continuity.

Important: do not run onboarding, reset, cleanup, restore, rm, mv, or tar extraction commands unless I explicitly approve after you explain the plan.

Then paste its answer back here.

ionic dew Apr 29, 2026, 1:21 AM

#

Here's the detailed check on your current setup:

Config File Path: Not explicitly mentioned in status output.
Workspace Path: /root/.openclaw/workspace.
Sessions Store Path: /root/.openclaw/agents/main/sessions/sessions.json.
Current Chat Session: Using a fresh session with ID: 65b10a1a-b0a6-4f2e-864b-0f7d94d81b2e.

Looking through previous sessions:

Old Main Session: It seems this current session is new. No older session was found explicitly related to our query.

Rebuilding Continuity:
Your MEMORY.md, AGENTS.md, and SOUL.md are sufficient to ensure continuity. They contain the needed guidance, settings, and information to resume smoothly.

If there's anything else you need to explore or validate, let me know!

Sent from my Galaxy

fringe spruceBOT Apr 29, 2026, 1:21 AM

#

Okay — that means the old live chat session probably did not restore, but the workspace/memory/personality files did.

One red flag: it still didn’t run/return openclaw config file; it just says “not explicitly mentioned.” So ask one final concrete thing:

Run exactly this and paste the raw output:

openclaw config file
openclaw sessions --all-agents --json | head -200

Do not summarize. Paste raw output only.

If the sessions list really has no old sessions, then there isn’t a session to switch back into. In that case, the practical fix is to continue from the restored MEMORY.md, AGENTS.md, and SOUL.md — that’s the durable “brain.” The exact short-term conversation state may be gone or was never included/restored.

Also: /root/.openclaw/... is fine only if your old install also ran as root. If your old backup came from /home/youruser/.openclaw/..., that mismatch matters.

ionic dew Apr 29, 2026, 1:22 AM

#

I want all my settings like discord bot all the things it set up for me

fringe spruceBOT Apr 29, 2026, 1:23 AM

#

Got it — then you’re not mainly looking for the old chat session. You want the old config + credentials + channel settings restored.

Paste this whole message to your bot:

I want my old OpenClaw settings restored: Discord bot setup, channel settings, provider/model settings, credentials/auth profiles, and anything onboarding configured.

Do not delete, reset, or overwrite anything yet.

Please check read-only:

1. Run `openclaw config file` and tell me the exact active config path.
2. Check whether the active config contains my old Discord settings under `channels.discord` / Discord accounts. Do not paste tokens or secrets.
3. Check whether model/provider settings and auth profiles exist.
4. Check whether the backup archive/restore source contains the old config, credentials directory, auth profiles, and Discord channel settings.
5. Compare current active config vs backup config and tell me what is missing.

#

6. If settings are missing, give me a safe restore plan first — do not apply it until I approve.

Important: do not paste Discord bot tokens, API keys, OAuth files, or secrets in chat. Only say whether they are present/missing.

The key thing: Discord bot setup lives in config/credentials, not just MEMORY.md or sessions. If onboarding made a fresh config, your bot may “remember” you but still be missing the actual Discord/provider settings.

Docs: https://docs.openclaw.ai/cli/backup

ionic dew Apr 29, 2026, 9:17 AM

#

I am having trouble configuring the back up so my not reminders it’s previous state

fringe spruceBOT Apr 29, 2026, 9:18 AM

#

That sounds like the backup was accepted, but the restored install isn’t pointing at the old state/session/reminder data.

First, please paste:

openclaw --version
openclaw backup verify /path/to/your-backup.tar.gz

Then check what’s actually inside the backup:

tar -tzf /path/to/your-backup.tar.gz | grep -E 'manifest.json|cron|sessions|memory|MEMORY|openclaw.json' | head -100

OpenClaw backups should include local state, config, auth profiles, sessions, and optionally workspaces; if the backup was made with --only-config or --no-include-workspace, it may not contain the old memory/workspace pieces you expected. (docs.openclaw.ai)

OpenClaw

Backup - OpenClaw

#

Also: don’t paste the full backup or auth files here. If you paste the command output, redact paths/tokens if needed. Happy to help interpret it 🦐

ionic dew Apr 29, 2026, 4:32 PM

#

I tried to change my open ai model during onboarding but it did not save

fringe spruceBOT Apr 29, 2026, 4:33 PM

#

Yep — onboarding can be finicky here, and existing sessions can also keep their own model override.

Please run:

openclaw --version
openclaw models status
openclaw models list --provider openai

If you’re using OpenAI Codex subscription/OAuth, also try:

openclaw models list --provider openai-codex

To set the default after onboarding, use the explicit model command:

#

openclaw models set <provider/model>
openclaw models status
openclaw gateway restart

Example shape only — pick the exact ref from models list:

openclaw models set openai/your-model-id

Also check the current Discord thread/session with:

/model status

#

If that session has its own override, set it in chat with:

/model <provider/model>

Docs: https://docs.openclaw.ai/cli/models and https://docs.openclaw.ai/concepts/models

Paste the openclaw --version + openclaw models status output here, but redact any keys/tokens.

ionic dew Apr 29, 2026, 4:33 PM

#

How do I change it

fringe spruceBOT Apr 29, 2026, 4:34 PM

#

From the machine running OpenClaw:

openclaw models list --provider openai
openclaw models set <provider/model-from-the-list>
openclaw models status
openclaw gateway restart

If you’re using OpenAI Codex / subscription OAuth, use:

openclaw models list --provider openai-codex
openclaw models set <provider/model-from-the-list>
openclaw gateway restart

#

For just the current Discord thread/session, type in chat:

/model

or:

/model <provider/model>

If it still doesn’t stick, paste:

openclaw --version

#

openclaw models status

Redact any keys/tokens. Docs: https://docs.openclaw.ai/cli/models

ionic dew Apr 29, 2026, 4:34 PM

#

Give me the steps one by one

fringe spruceBOT Apr 29, 2026, 4:34 PM

#

Sure — do these one by one.

1. Open a terminal on the machine running OpenClaw.

2. Check your version:

openclaw --version

3. List available OpenAI models:

openclaw models list --provider openai

If you use OpenAI Codex subscription/OAuth, run this instead:

#

openclaw models list --provider openai-codex

4. Copy the exact model ref from the list.
It should look like:

provider/model-name

5. Set it as your default:

openclaw models set provider/model-name

Replace provider/model-name with the exact one you copied.

#

6. Confirm it saved:

openclaw models status

7. Restart OpenClaw:

openclaw gateway restart

8. If this Discord thread still uses the old model, type here:

/model status

Then set the thread/session too:

#

/model provider/model-name

Docs: https://docs.openclaw.ai/cli/models

ionic dew Apr 29, 2026, 4:35 PM

#

How do I save it

fringe spruceBOT Apr 29, 2026, 4:36 PM

#

To save the model as the default, run this:

openclaw models set provider/model-name

Replace provider/model-name with the exact model you picked from the list.

Example shape:

openclaw models set openai/some-model-name

Then confirm it saved:

#

openclaw models status

Then restart the gateway so the running bot picks it up:

openclaw gateway restart

For this current Discord thread only, you can save a session override by typing:

/model provider/model-name

The CLI one is the important “default save.” Docs: https://docs.openclaw.ai/cli/models

ionic dew Apr 29, 2026, 6:39 PM

#

Can I install got codex 5.1 by running onboarding again

fringe spruceBOT Apr 29, 2026, 6:39 PM

#

Yes, but I’d not run full onboarding again first unless you want to reconfigure everything.

For Codex/OpenAI Codex, do this instead:

1. Check version

openclaw --version

2. Login/configure Codex auth

openclaw models auth login --provider openai-codex --set-default

3. See available Codex models

#

openclaw models list --provider openai-codex

4. Set the exact model you want

openclaw models set openai-codex/model-name-from-list

5. Confirm

openclaw models status

6. Restart

#

openclaw gateway restart

If you mean a specific “Codex 5.1” model, pick the exact name from models list — model names change, so I don’t want to guess. For latest model recommendations, #models is the best place.

Docs: https://docs.openclaw.ai/cli/models

ionic dew Apr 29, 2026, 6:40 PM

#

I have seen the list it’s not there I want to add it during onboarding the model I want is there

fringe spruceBOT Apr 29, 2026, 6:41 PM

#

Yes — if you see it in onboarding, use the model configuration wizard instead of full reset.

Do this:

1. Run:

openclaw configure --section model

2. When asked about existing config, choose Modify / keep existing config.
Do not choose Reset.

3. Choose the OpenAI/Codex provider option you use.

4. In the model picker, select the Codex model you want.

5. If there is an allowlist / multi-select screen, make sure that model is checked there too.

#

This is what makes it appear in /model.

6. Finish the wizard fully. Don’t quit halfway.

7. Confirm it saved:

openclaw models status

8. Restart:

openclaw gateway restart

9. In this Discord thread, if it still uses the old model, type:

#

/model

and select the new one for this session too.

Docs: https://docs.openclaw.ai/cli/configure

ionic dew Apr 29, 2026, 6:46 PM

#

It only lists two models I want to add 1

fringe spruceBOT Apr 29, 2026, 6:46 PM

#

That usually means your model allowlist only has two models. Add the one you want to agents.defaults.models, then set it.

1. Get the exact model ref
It must look like:

provider/model-name

Example provider prefixes are usually openai/... or openai-codex/....

2. Add it to the allowed models list:

openclaw config set agents.defaults.models '{"provider/model-name":{}}' --strict-json --merge

Replace provider/model-name with the exact model.

#

3. Set it as default:

openclaw models set provider/model-name

4. Confirm:

openclaw models status

5. Restart:

openclaw gateway restart

If you paste the exact model name you’re trying to add, I can format the command for you. Docs: https://docs.openclaw.ai/concepts/models

ionic dew Apr 29, 2026, 6:46 PM

#

Can I add during onboarding

fringe spruceBOT Apr 29, 2026, 6:47 PM

#

Yes — but use configure, not a full onboarding reset.

Run:

openclaw configure --section model

Then:

Choose Modify / keep existing config
Choose your provider, e.g. OpenAI or OpenAI Codex
In the model picker / allowlist step, select the model you want to add
Finish the wizard completely
Restart:

#

openclaw gateway restart

If it still only shows two models, add it directly with:

openclaw config set agents.defaults.models '{"provider/model-name":{}}' --strict-json --merge
openclaw models set provider/model-name
openclaw gateway restart

Docs: https://docs.openclaw.ai/cli/configure

ionic dew Apr 29, 2026, 6:47 PM

#

Model allowlist does not show the model I want

fringe spruceBOT Apr 29, 2026, 6:48 PM

#

If the allowlist picker doesn’t show it, add it manually.

1. First get your version:

openclaw --version

2. Try the full provider catalog:

openclaw models list --all --provider openai-codex

or, if it’s normal OpenAI API:

openclaw models list --all --provider openai

#

3. Copy the exact model ref.
It must look like:

provider/model-name

4. Add it to the allowlist:

openclaw config set agents.defaults.models '{"provider/model-name":{}}' --strict-json --merge

5. Set it as default:

openclaw models set provider/model-name

#

6. Restart:

openclaw gateway restart

If you tell me the exact model name you want, I’ll write the exact command for you. Docs: https://docs.openclaw.ai/concepts/models

ionic dew Apr 29, 2026, 6:48 PM

#

Can’t I add at onboarding

fringe spruceBOT Apr 29, 2026, 6:48 PM

#

Yes, you can — but don’t reset onboarding.

Run:

openclaw onboard

Then:

When it detects existing config, choose Modify / Keep existing
Go to Model and auth
Pick OpenAI Codex / OpenAI Code subscription if that’s what you use
If it offers manual model entry, type the exact model ref, like:

#

openai-codex/model-name

or:

openai/model-name

Finish onboarding completely
Restart:

openclaw gateway restart

#

If onboarding still won’t show or accept it, then you’ll need to add it manually with openclaw config set. The picker only shows models OpenClaw knows about. Docs: https://docs.openclaw.ai/start/wizard-cli-reference

ionic dew Apr 29, 2026, 6:50 PM

#

Configure handling what should I choose

fringe spruceBOT Apr 29, 2026, 6:50 PM

#

Choose Modify existing config / Keep existing config.

Do not choose Reset unless you want to wipe/recreate config.

Then in configure:

Choose Model
Choose your provider:
- OpenAI Codex / OpenAI Code subscription if you use Codex subscription/OAuth
- OpenAI API key if you use an API key
In the model allowlist picker, select the model you want
Finish the wizard fully
Restart:

openclaw gateway restart

#

If the model still doesn’t appear in the picker, the wizard can’t see it from the catalog, so you’ll need to add it manually with openclaw config set.

ionic dew Apr 29, 2026, 6:51 PM

#

Use existing values or update?

fringe spruceBOT Apr 29, 2026, 6:51 PM

#

Choose Update for the model section.

Use the existing values for anything unrelated, like gateway, channels, workspace, Discord token, etc.

So basically:

Model/auth: Update
Workspace: Use existing
Gateway: Use existing
Channels/Discord: Use existing
Daemon/service: Use existing unless it asks to restart/reinstall
Reset: No / don’t reset

After finishing:

#

openclaw gateway restart

ionic dew Apr 29, 2026, 6:51 PM

#

This is configure handling stage

fringe spruceBOT Apr 29, 2026, 6:52 PM

#

At the configure handling stage, choose:

Update

That lets you change the model settings.

Do not choose reset/wipe.

After that:

Go to Model/auth
Pick your OpenAI/OpenAI Codex option
Add/select the model
Finish configure completely
Restart:

#

openclaw gateway restart

ionic dew Apr 29, 2026, 6:55 PM

#

How can I see if my model changed

fringe spruceBOT Apr 29, 2026, 6:56 PM

#

Check it two ways:

1. Terminal default model

openclaw models status

Look for the primary/default model line.

2. Current Discord session model
Type here in Discord:

/model status

If those differ, the Discord thread has a session override. To force this thread to the new model:

#

/model provider/model-name

After changing the default, also restart:

openclaw gateway restart

ionic dew Apr 29, 2026, 6:56 PM

#

What does hatch terminal do

fringe spruceBOT Apr 29, 2026, 6:56 PM

#

I think you mean chat terminal / TUI.

It opens a terminal chat window where you can talk to your OpenClaw directly from the machine, instead of Discord.

It does not restore backups or add models by itself. For your model issue, you don’t need it.

Use it only if you want to test locally:

openclaw chat

or:

openclaw tui

#

To check your model, keep using:

openclaw models status

Docs: https://docs.openclaw.ai/cli/tui

ionic dew Apr 29, 2026, 6:57 PM

#

Hatch in terminal

fringe spruceBOT Apr 29, 2026, 6:58 PM

#

“Hatch in terminal” means open a terminal chat/test session with your bot.

It usually sends the first “wake up” message to check that:

Gateway is running
Auth works
The model can reply
Your local terminal can talk to OpenClaw

It does not add or save models by itself.

If you’re at the end of setup/configure, you can choose it to test. If it hangs or no reply comes back, run:

openclaw gateway restart
openclaw status

#

openclaw models status
openclaw logs --follow

Then if needed:

openclaw doctor

Docs note that the TUI sends “Wake up, my friend!” on first hatch, and if it gets no reply/tokens stay at 0, the agent never ran. (docs.openclaw.ai)

ionic dew Apr 29, 2026, 7:01 PM

#

I have changed the model but it still says the old one

fringe spruceBOT Apr 29, 2026, 7:01 PM

#

That usually means you added it to the allowlist, but didn’t set it as the default, or this Discord thread has its own old session override.

Do this:

1. Check what the default is:

openclaw models status --plain

2. Force the default to the new model:

openclaw models set provider/model-name

Use the exact model ref you added.

3. Restart:

#

openclaw gateway restart

4. Now check this Discord thread/session:

/model status

If that still shows the old model, set this thread too:

/model provider/model-name

If it still refuses to change, paste the output of:

#

openclaw --version
openclaw models status

and the result of:

/model status

Redact any keys/tokens.

ionic dew Apr 29, 2026, 7:02 PM

#

How do I exit chat terminal

fringe spruceBOT Apr 29, 2026, 7:03 PM

#

Press:

Ctrl+C

If that doesn’t exit, try:

Ctrl+D

or type:

/exit

#

Then you’ll be back at your normal terminal prompt.

ionic dew Apr 29, 2026, 7:08 PM

#

Terminal is just stuck

#

I can’t close

fringe spruceBOT Apr 29, 2026, 7:08 PM

#

If it’s stuck in chat terminal / hatch, that’s safe to exit.

Try in order:

1. Press Enter once

2. Press:

Ctrl+C

3. If still stuck, press:

Ctrl+\

4. If that still doesn’t work, close that terminal window and open a new one.

#

Then in the new terminal run:

openclaw status

If OpenClaw looks weird after that:

openclaw gateway restart

Don’t choose reset or wipe anything.

#

No worries — don’t reset anything.

Try this order:

1. Click inside the stuck terminal and press:

Ctrl+C

2. If that fails:

Ctrl+\

3. If it still won’t close, force-close the terminal app:

Mac: Cmd+Q, or Apple menu → Force Quit → Terminal

#

Windows: Ctrl+Shift+Esc → Task Manager → end Terminal/PowerShell
Linux: close the terminal window, or use your system monitor

Then open a fresh terminal and run:

openclaw status

If needed after that:

openclaw gateway restart

ionic dew Apr 29, 2026, 7:11 PM

#

How do I check my model allow list

fringe spruceBOT Apr 29, 2026, 7:11 PM

#

Run this in terminal:

openclaw config get agents.defaults.models

That shows your model allowlist.

For a cleaner full model check, also run:

openclaw models status

If the allowlist is empty/missing, OpenClaw may use the broader configured/catalog models instead.

To see what Discord can pick from, type here:

#

/model list

or:

/models

ionic dew Apr 29, 2026, 7:13 PM

#

Ok it lists 3 how do I make sure codex is the main

fringe spruceBOT Apr 29, 2026, 7:13 PM

#

Set the Codex model as the primary/default.

1. Copy the exact Codex ref from your list
It should look like:

openai-codex/...

2. Run:

openclaw models set openai-codex/your-model-name

3. Confirm:

#

openclaw models status --plain

It should print the Codex model.

4. Restart:

openclaw gateway restart

5. If this Discord thread still shows the old model, type here:

/model openai-codex/your-model-name

Then check:

#

/model status

ionic dew Apr 29, 2026, 7:14 PM

#

No it does not

fringe spruceBOT Apr 29, 2026, 7:14 PM

#

Got it — then force it directly in config.

1. Copy the exact Codex model from your allowlist.

2. Run this, replacing the model name:

openclaw config set agents.defaults.model '{"primary":"openai-codex/your-model-name"}' --strict-json

3. Check it:

openclaw config get agents.defaults.model
openclaw models status --plain

4. Restart:

#

openclaw gateway restart

5. For this Discord thread, set it too:

/model openai-codex/your-model-name

If that still shows old, paste the output of:

openclaw config get agents.defaults.models
openclaw config get agents.defaults.model
openclaw models status

#

Redact tokens only.

ionic dew Apr 29, 2026, 7:14 PM

#

How do I select it to primary

fringe spruceBOT Apr 29, 2026, 7:15 PM

#

Do this:

1. Show your allowlist:

openclaw config get agents.defaults.models

2. Copy the exact Codex model name from that output. It should be like:

openai-codex/...

3. Set it as primary:

openclaw models set openai-codex/your-model-name

#

4. Check it saved:

openclaw models status --plain

5. Restart OpenClaw:

openclaw gateway restart

If step 3 doesn’t change it, force primary directly:

openclaw config set agents.defaults.model '{"primary":"openai-codex/your-model-name"}' --strict-json
openclaw gateway restart

ionic dew Apr 29, 2026, 7:15 PM

#

Can’t I do it in the config file

fringe spruceBOT Apr 29, 2026, 7:15 PM

#

Yes. You can edit the config file directly.

1. Find the config file:

openclaw config file

2. Open that file in your editor.

3. Find this area:

agents: {
  defaults: {

4. Set the primary model like this:

#

agents: {
  defaults: {
    model: {
      primary: "openai-codex/your-model-name",
    },
  },
}

If you already have model: there, replace just that part.

5. Make sure the same model is also in the allowlist:

models: {
  "openai-codex/your-model-name": {},

#

inside:

agents: {
  defaults: {
    models: {
      "openai-codex/your-model-name": {},
    },
  },
}

6. Save the file.

#

7. Validate it:

openclaw config validate

8. Restart:

openclaw gateway restart

If you paste the agents.defaults section with tokens removed, I can point to the exact line.

ionic dew Apr 29, 2026, 7:16 PM

#

How do I get the file to edit

fringe spruceBOT Apr 29, 2026, 7:16 PM

#

Run this:

openclaw config file

It will print the path to your config file, usually something like:

/home/you/.openclaw/openclaw.json

Then open it with an editor.

If you’re on Mac/Linux:

#

nano "$(openclaw config file)"

Save in nano with:

Ctrl+O
Enter
Ctrl+X

If you’re on Windows/WSL:

notepad "$(openclaw config file)"

#

After editing, run:

openclaw config validate
openclaw gateway restart

ionic dew Apr 29, 2026, 7:18 PM

#

Ok in that file codex is primary

fringe spruceBOT Apr 29, 2026, 7:18 PM

#

Good — then the config default is probably correct. If it still says the old model, it’s likely one of these:

Gateway hasn’t restarted
This Discord thread has a session override
A different agent/per-agent config is overriding the default

Do this:

1. Validate + restart

openclaw config validate
openclaw gateway restart

2. Confirm terminal sees Codex

#

openclaw models status --plain

3. In this Discord thread, check session override

/model status

If it shows old model, set this thread manually:

/model openai-codex/your-model-name

4. If it still shows old, check per-agent overrides:

#

openclaw config get agents.list

Look for any model: under an agent. That can override agents.defaults.model.

ionic dew Apr 29, 2026, 9:05 PM

#

I am using a different model for WhatsApp but I want use the same model as my main

fringe spruceBOT Apr 29, 2026, 9:05 PM

#

Yes — WhatsApp is probably routed to a different agent with its own model.

Do this:

1. Check your main model

openclaw models status --plain

Copy the model it prints.

2. Check which agent WhatsApp uses

openclaw agents list --bindings

Look for the WhatsApp binding and note the agentId.

#

3. Check agent overrides

openclaw config get agents.list

Find that WhatsApp agent. If it has a model: line, that overrides your main/default model.

4. To make WhatsApp use the main model:
Open config:

nano "$(openclaw config file)"

Find the WhatsApp agent under agents.list, then either:

remove its model: entry so it inherits agents.defaults.model, or

#

set it to the same model as main.

5. Save, validate, restart

openclaw config validate
openclaw gateway restart

Also in your WhatsApp chat, send:

/model status

If it still shows the old model, send:

#

/model your-main-model

Docs: https://docs.openclaw.ai/concepts/multi-agent

ionic dew Apr 29, 2026, 9:12 PM

#

Agents:

main (default)
Identity: 🎧 AI DJ (IDENTITY.md)
Workspace: ~/.openclaw/workspace
Agent dir: ~/.openclaw/agents/main/agent
Model: openai/gpt-5.1
Routing rules: 0
Routing: default (no explicit rules)
Providers:
- WhatsApp default: linked
  Routing rules map channel/account/peer to an agent. Use --bindings for full rules.
  Channel status reflects local config/creds. For live health: openclaw channels status --probe.
  root@OpenClaw:~#

fringe spruceBOT Apr 29, 2026, 9:12 PM

#

That output means WhatsApp is already routed to main. There is no separate WhatsApp agent.

So if WhatsApp still shows a different model, it’s almost certainly a WhatsApp session override.

In your WhatsApp chat, send:

/model status

If it shows the wrong model, send:

/model openai/gpt-5.1

Then check again:

#

/model status

If your real “main” model is supposed to be Codex, then this output shows it is not Codex right now — it says:

Model: openai/gpt-5.1

So either set Codex as main, or set WhatsApp to openai/gpt-5.1 to match what main currently is.

ionic dew Apr 29, 2026, 9:13 PM

#

What model is WhatsApp using

fringe spruceBOT Apr 29, 2026, 9:13 PM

#

From the output you pasted: WhatsApp is routed to main, and main is using:

openai/gpt-5.1

So WhatsApp should be using openai/gpt-5.1 by default.

To confirm inside WhatsApp, send this to your bot in WhatsApp:

/model status

That will show if the WhatsApp session has an override.

ionic dew Apr 29, 2026, 9:14 PM

#

With the slash

fringe spruceBOT Apr 29, 2026, 9:14 PM

#

Yes — send it exactly like this in WhatsApp:

/model status

It needs the slash at the start.

If it shows the wrong one, send:

/model openai/gpt-5.1

ionic dew Apr 29, 2026, 9:15 PM

#

It’s not even working again

fringe spruceBOT Apr 29, 2026, 9:15 PM

#

Ugh, okay — let’s stop chasing model settings and check whether WhatsApp is actually connected.

Run these on the server:

openclaw status
openclaw channels status --probe
openclaw gateway status

Then restart once:

openclaw gateway restart

Wait ~20 seconds, then run:

#

openclaw channels status --probe

If WhatsApp still doesn’t reply, check logs:

openclaw logs --follow

Then send a WhatsApp message to the bot and watch for errors.

Paste the output of:

openclaw --version

#

openclaw channels status --probe

Redact tokens/phone numbers if they show.

ionic dew Apr 29, 2026, 9:19 PM

#

WhatsApp default: enabled, configured, linked, stopped, disconnected, dm:allowlist, allow:

Tip: status --deep adds gateway health probes to status output (requires a reachable gateway).
root@OpenClaw:~#

fringe spruceBOT Apr 29, 2026, 9:19 PM

#

That line explains it:

linked, stopped, disconnected

So the WhatsApp account exists, but the WhatsApp listener is not running/connected.

Do this:

1. Restart the gateway

openclaw gateway restart

2. Wait 20–30 seconds

#

3. Check again

openclaw channels status --probe

You want WhatsApp to say something like running/connected.

If it still says stopped/disconnected, run:

openclaw logs --follow

Then send a WhatsApp message to the bot and watch the logs. Paste the WhatsApp-related error here.

Also check:

#

openclaw doctor

Likely issue is the gateway didn’t start the WhatsApp plugin/listener after the config changes.

ionic dew Apr 29, 2026, 9:22 PM

#

Ok it’s connected now

fringe spruceBOT Apr 29, 2026, 9:23 PM

#

Nice, good — that was the blocker 🎉

Now test WhatsApp directly:

/model status

If it shows the wrong model, send:

/model openai/gpt-5.1

Then send a normal test message like:

#

hello

If it replies, you’re back in business. If this fully fixed it, please right-click the solution message → Apps → ✅ Mark Solution so Answer Overflow indexes it for future users.

ionic dew Apr 29, 2026, 9:24 PM

#

But I just sent WhatsApp a message and it’s not replying

fringe spruceBOT Apr 29, 2026, 9:24 PM

#

The clue is your earlier status showed:

dm:allowlist, allow:

That means WhatsApp DMs are in allowlist mode, but the allowlist looks empty. So it may receive the message but refuse to answer.

Do this:

1. Add your WhatsApp number to the allowlist
Use your real number in international format:

openclaw config set channels.whatsapp.allowFrom '["+1234567890"]' --strict-json

#

2. Validate

openclaw config validate

3. Restart

openclaw gateway restart

4. Check

openclaw channels status --probe

Then send WhatsApp a normal message again.

#

If it still doesn’t reply, run this and send another WhatsApp message while watching:

openclaw logs --follow

Also please paste:

openclaw --version

Docs: https://docs.openclaw.ai/channels/whatsapp

ionic dew Apr 29, 2026, 9:25 PM

#

It’s not empty I did not want to put my number here

fringe spruceBOT Apr 29, 2026, 9:25 PM

#

Totally fair — don’t paste your number here. I shouldn’t have implied you needed to.

Since it’s connected and allowlist is configured, check whether OpenClaw is seeing the inbound WhatsApp message:

openclaw logs --follow

Then send one WhatsApp message. Look for either:

blocked / unauthorized / allowlist
model/auth error
no inbound log at all

Also run:

#

openclaw channels status --probe
openclaw status --deep

If you paste logs, redact phone numbers/tokens. The key thing is the error wording around the WhatsApp message.

ionic dew Apr 29, 2026, 9:30 PM

#

Tip: status --deep adds gateway health probes to status output (requires a reachable gateway).
root@
◇
Gateway reachable.

WhatsApp default: enabled, configured, linked, running, disconnected, in:3m ago, dm:allowlist, allow:error:status=408 Request Time-out Connection was lost

Tip: status --deep adds gateway health probes to status output (requires a reachable gateway).
root@OpenClaw:~#

fringe spruceBOT Apr 29, 2026, 9:30 PM

#

That’s not a model problem now — WhatsApp is still not fully connected:

running, disconnected
allow:error:status=408 Request Time-out Connection was lost

Do this next:

1. Restart and wait

openclaw gateway restart

Wait 60 seconds, then:

#

openclaw channels status --probe

2. If it still says disconnected, re-link WhatsApp

openclaw channels login --channel whatsapp --account default

Scan the QR if it gives you one.

3. Restart again

openclaw gateway restart

4. Check

#

openclaw channels status --probe

You want it to say running, connected.

If it keeps showing 408 Request Time-out, check logs:

openclaw logs --follow

A 408 often means the WhatsApp Web connection dropped/timed out — sometimes proxy/network/VPS related. Docs: https://docs.openclaw.ai/channels/troubleshooting

ionic dew Apr 29, 2026, 9:34 PM

#

Still says disconnected

fringe spruceBOT Apr 29, 2026, 9:34 PM

#

Okay — then do the re-link step now.

1. First check version:

openclaw --version

2. Re-link WhatsApp:

openclaw channels login --channel whatsapp --account default

Scan the QR if it shows one.

3. Restart gateway:

#

openclaw gateway restart

4. Wait 60 seconds, then check:

openclaw channels status --probe

You want WhatsApp to say running, connected.

If it still says disconnected, run:

openclaw logs --follow

#

Then paste the WhatsApp error lines here — redact phone numbers/tokens. This is likely a WhatsApp Web session/network timeout now, not the model.

ionic dew Apr 29, 2026, 9:47 PM

#

Openclaw 5.5 good for coding

fringe spruceBOT Apr 29, 2026, 9:47 PM

#

If you mean GPT/OpenAI 5.5, yes — it should be strong for coding.

But I don’t want to give stale model rankings here. For current “best coding model” advice, check #models — people there have fresher hands-on results.

Docs for model setup: https://docs.openclaw.ai/concepts/models

ionic dew Apr 29, 2026, 9:58 PM

#

WhatsApp looks like it wants to talk then stops

fringe spruceBOT Apr 29, 2026, 9:59 PM

#

That sounds like OpenClaw is receiving the WhatsApp message and starting a run, then failing during model/auth or send-back.

Run this and watch while you send one WhatsApp message:

openclaw logs --follow

Then paste the error lines around that message — redact phone numbers/tokens.

Also run:

openclaw status --deep
openclaw models status
openclaw channels status --probe

#

Common causes at this stage:

WhatsApp transport keeps disconnecting
sender is blocked by allowlist/pairing
model auth fails after the message is accepted
selected model is not allowed / not reachable
outbound WhatsApp send fails

The logs will tell us which one.

ionic dew Apr 29, 2026, 10:02 PM

#

Gateway keeps timing out

fringe spruceBOT Apr 29, 2026, 10:02 PM

#

Okay — pause the WhatsApp testing for a sec. Let’s diagnose the gateway itself.

Run these in a fresh terminal:

openclaw --version
openclaw gateway status
openclaw status --all

Then restart cleanly:

openclaw gateway restart

Wait 60 seconds, then:

#

openclaw status --deep
openclaw channels status --probe

If it still times out, watch logs:

openclaw logs --follow

Then send one WhatsApp message and copy only the error lines around it.

Do not reset/reinstall yet. This sounds like either:

gateway is hanging/crashing,
WhatsApp Web transport is timing out,
or the model run starts then fails.

#

Paste openclaw --version, gateway status, and the timeout/error line — redact phone numbers/tokens.

ionic dew Apr 29, 2026, 10:09 PM

#

fringe spruce Common causes at this stage: - WhatsApp transport keeps disconnecting - sender ...

No something went wrong while processing your request

fringe spruceBOT Apr 29, 2026, 10:09 PM

#

That message means WhatsApp is reaching OpenClaw, but the agent run is failing after the message is accepted. So don’t relink again yet.

Run:

openclaw --version
openclaw models status
openclaw channels status --probe

Then watch logs:

openclaw logs --follow

While logs are open, send one WhatsApp message. Look for the first error after that.

#

Most likely causes now:

model auth failed
openai/gpt-5.1 is not allowed/reachable
session has old model override
outbound WhatsApp send failed

If you want to test model auth directly, run:

openclaw models status --probe

That may make a small live provider request.

Paste the error lines from logs here, but redact phone numbers/tokens/API keys.

ionic dew Apr 30, 2026, 12:54 AM

#

It seams I have things up and running again last thing I want to use different models for different things how do I configure or do I just tell my bot through WhatsApp basically I want to use a cheap but good model as a dj customer assistant then when I want to program things a good model that can debug run script and code I want tho set up to be cost effective as at the moment I am spending too much a month

fringe spruceBOT Apr 30, 2026, 12:54 AM

#

Nice — glad it’s mostly back up.

For what you want, I’d do both:

Simple setup: one bot, switch models when needed

Set your cheap model as the default/main for WhatsApp customer-assistant work.

Then when you want coding/debugging, send in WhatsApp:

/model

or:

#

/model provider/coding-model

When done coding, switch back:

/model provider/cheap-model

This is easiest, but you have to remember to switch back.

Better cost-control setup: separate agents

Recommended:

main / WhatsApp DJ assistant → cheap-but-good model

#

coding agent → stronger coding model, used only when programming

Steps:

openclaw agents add coding

Then configure that agent with the stronger model, and keep main on the cheaper model.

Check routing:

openclaw agents list --bindings

If you want WhatsApp to stay cheap, leave WhatsApp routed to main.

#

Use the coding agent through another surface like terminal/chat, Control UI, Telegram, or a separate Discord thread so it doesn’t burn expensive tokens for customer chat.

Key idea

Don’t use your best coding model as the everyday WhatsApp assistant. Make the cheap model default, and only switch/escalate when you’re doing code.

For current “best cheap model” recommendations, ask in #models — model pricing/performance changes too fast for me to confidently rank them here.

Docs:
https://docs.openclaw.ai/concepts/models
https://docs.openclaw.ai/concepts/multi-agent

ionic dew Apr 30, 2026, 12:56 AM

#

I am not going to rennet to switch back and the customer service agent will be used by the public as well so how can I config

fringe spruceBOT Apr 30, 2026, 12:56 AM

#

Don’t rely on remembering to switch. Configure separate agents.

Recommended setup:

customer agent → cheap model, public-safe, limited tools
coding agent → expensive/strong coding model, only you use it

Since your WhatsApp is public/customer-facing, keep WhatsApp routed to the cheap customer agent.

Example config shape in openclaw.json:

{
  agents: {
    defaults: {
      models: {

#

        "provider/cheap-model": {},
        "provider/coding-model": {},
      },
    },
    list: [
      {
        id: "customer",
        name: "DJ Customer Assistant",
        workspace: "~/.openclaw/workspace-customer",
        model: "provider/cheap-model",
        tools: {
          deny: ["exec", "write", "edit", "apply_patch", "browser", "message"]
        }
      },
      {

#

        id: "coding",
        name: "Coding Assistant",
        workspace: "~/.openclaw/workspace-coding",
        model: "provider/coding-model"
      }
    ]
  },

  bindings: [
    { agentId: "customer", match: { channel: "whatsapp" } }
  ]
}

Important: for public WhatsApp, also make sure only you can run slash/model commands, otherwise customers may be able to change behavior. Set command access for just your number, not the public.

#

Then:

openclaw config validate
openclaw gateway restart
openclaw agents list --bindings

Use the coding agent from Control UI / terminal / another private channel, not the public WhatsApp.

Docs: https://docs.openclaw.ai/concepts/multi-agent https://docs.openclaw.ai/channels/channel-routing

ionic dew Apr 30, 2026, 12:58 AM

#

No I mean is there a way that when it is a simple question it routes to one model then when it has to do with coding it roots to another

fringe spruceBOT Apr 30, 2026, 12:58 AM

#

Short answer: not automatically by “simple vs coding” from the same chat, at least not as a simple config switch.

OpenClaw routing is mostly based on channel/account/thread/peer → agent, not message meaning.

Best safe setup:

WhatsApp public customer bot
- cheap model
- limited/no dangerous tools
- no public /model switching
Private coding agent
- strong coding model
- exec/scripts/debug tools enabled
- use from Control UI, terminal, Discord private thread, Telegram, etc.

So your config should route WhatsApp → customer agent always. Then use a separate private coding route when you want coding.

#

Do not rely on fallbacks for this — fallbacks are for model failure, not “hard question detected.”

If you really want automatic content-based routing, that would need a custom workflow/plugin/agent instruction, but I wouldn’t put that on public WhatsApp because customers could accidentally or intentionally trigger the expensive coding model.

ionic dew Apr 30, 2026, 12:59 AM

#

No only I have access to WhatsApp

fringe spruceBOT Apr 30, 2026, 1:00 AM

#

Ah, got it — if only you use WhatsApp, then you can make this much simpler.

Best practical setup:

Set cheap model as the default for WhatsApp everyday questions.
When you want coding, explicitly start a coding session/agent from WhatsApp.

Automatic “detect coding → route to coding model” isn’t really the normal routing model. But you can make it easy by creating a coding agent and using it intentionally.

Config idea:

agents: {
  list: [
    {
      id: "main",

#

      name: "Everyday DJ Assistant",
      model: "provider/cheap-model",
      workspace: "~/.openclaw/workspace",
    },
    {
      id: "coding",
      name: "Coding Assistant",
      model: "provider/coding-model",
      workspace: "~/.openclaw/workspace-coding",
    },
  ],
}

Then keep WhatsApp routed to main.

#

When you need coding, use a clear command/pattern like:

/focus coding

or use a separate private thread/channel bound to the coding agent, depending on how your setup is configured.

If you want one-command switching inside WhatsApp, /model provider/coding-model works — but that changes the session until switched back, so I’d prefer a separate coding agent/session if cost matters.

twin matrixBOT Apr 30, 2026, 1:01 AM

#

ionic dew Apr 30, 2026, 1:01 AM

#

Ok I want to use a basic customer service agent when people ask questions on discord but I will only be asking complex questions via WhatsApp

fringe spruceBOT Apr 30, 2026, 1:01 AM

#

Yes — that’s the right split.

Set it up like this:

Discord → customer agent → cheap/basic model
WhatsApp → main/coding agent → stronger model

In openclaw.json, you want something like:

agents: {
  list: [
    {
      id: "main",
      default: true,
      name: "Main / Coding",

#

      model: "provider/strong-coding-model",
      workspace: "~/.openclaw/workspace",
    },
    {
      id: "customer",
      name: "Customer Service",
      model: "provider/cheap-customer-model",
      workspace: "~/.openclaw/workspace-customer",
      tools: {
        deny: ["exec", "write", "edit", "apply_patch", "browser"]
      }
    },
  ],
},

#

bindings: [
  { agentId: "customer", match: { channel: "discord" } },
  { agentId: "main", match: { channel: "whatsapp" } },
]

Then run:

openclaw config validate
openclaw gateway restart
openclaw agents list --bindings

Important: if Discord is public/customer-facing, keep dangerous tools off for that agent and don’t let public users run /model or config commands.

#

Docs: https://docs.openclaw.ai/concepts/multi-agent and https://docs.openclaw.ai/channels/channel-routing

ionic dew Apr 30, 2026, 1:05 AM

#

What was the switch method you told me about

fringe spruceBOT Apr 30, 2026, 1:06 AM

#

The switch method was the /model command.

In the chat where you want to change model, send:

/model status

to see current model.

Then switch with:

/model provider/model-name

Example:

#

/model openai/gpt-5.1

To see available models:

/model list

Important: this changes the current session/chat until you switch again. For your use case, better is still:

Discord → cheap customer agent via config bindings
WhatsApp → strong/main agent via config bindings

Then you don’t need to remember to switch.

ionic dew Apr 30, 2026, 1:07 AM

#

Can I tell my bot via WhatsApp to use different models like in discord

fringe spruceBOT Apr 30, 2026, 1:07 AM

#

Yes — if slash commands are enabled for WhatsApp and your sender is authorized, you can use the same text commands.

In WhatsApp send:

/model status

List options:

/model list

Switch model:

#

/model provider/model-name

Example:

/model openai/gpt-5.1

But remember: it changes that WhatsApp session until you switch again. For your Discord customer service vs WhatsApp coding split, config bindings are safer/cheaper.

ionic dew Apr 30, 2026, 1:08 AM

#

What are config bindings

fringe spruceBOT Apr 30, 2026, 1:08 AM

#

Config bindings are routing rules in openclaw.json.

They tell OpenClaw:

“Messages from this channel/account/thread should go to this agent.”

Example:

bindings: [
  { agentId: "customer", match: { channel: "discord" } },
  { agentId: "main", match: { channel: "whatsapp" } },
]

That means:

#

Discord messages use the customer agent/model
WhatsApp messages use the main agent/model

So instead of manually switching models, OpenClaw automatically picks the right agent based on where the message came from.

Check your current bindings with:

openclaw agents list --bindings

Docs: https://docs.openclaw.ai/channels/channel-routing

ionic dew Apr 30, 2026, 1:09 AM

#

I only have one agent

fringe spruceBOT Apr 30, 2026, 1:09 AM

#

Right — with one agent, bindings won’t help much because every channel goes to the same agent/model.

You have two choices:

Option A — keep one agent, switch manually

Use in WhatsApp/Discord:

/model list
/model provider/model-name

Simple, but you have to remember.

Option B — add a second agent

This is what you want for automatic Discord vs WhatsApp model split.

#

Run:

openclaw agents add customer

Set customer to your cheap customer-service model.

Then route Discord to customer, and keep WhatsApp on main.

Your config ends up like:

bindings: [
  { agentId: "customer", match: { channel: "discord" } },
  { agentId: "main", match: { channel: "whatsapp" } },

#

Then:

openclaw config validate
openclaw gateway restart
openclaw agents list --bindings

For your goal, I’d choose Option B.

ionic dew Apr 30, 2026, 1:10 AM

#

I is 5.1 exspensive

fringe spruceBOT Apr 30, 2026, 1:10 AM

#

Yes — GPT‑5.1 is relatively expensive for everyday customer-service chat, especially because output tokens cost much more than input tokens.

Official OpenAI pricing shows GPT‑5.1 / GPT‑5 pricing at about $1.25 per 1M input tokens and $10.00 per 1M output tokens. OpenAI also says GPT‑5.1 pricing/rate limits are the same as GPT‑5. (developers.openai.com)

So I’d use it like this:

Discord customer service: cheaper model
WhatsApp private coding/debugging: GPT‑5.1 / stronger model

Don’t use 5.1 as the default public customer-service model if you’re trying to reduce monthly spend. For current cheap model recommendations, ask in #models.

GPT-5.1 Model | OpenAI API

ionic dew Apr 30, 2026, 1:12 AM

#

I told my my WhatsApp channel gone down again I asked the bot to do something a bit complex and it has not replied me in like an hour but I can see he has configured some of what I asked

fringe spruceBOT Apr 30, 2026, 1:12 AM

#

Sounds like the agent run got stuck, not necessarily that WhatsApp is broken.

Do this:

1. Check current state

openclaw --version
openclaw status
openclaw channels status --probe

2. If it shows an active/stuck run, try stopping from WhatsApp:

/stop

3. If WhatsApp won’t respond, restart the gateway:

#

openclaw gateway restart

4. Wait 30–60 seconds, then check:

openclaw channels status --probe

5. If it still doesn’t reply, watch logs while sending one WhatsApp message:

openclaw logs --follow

Since it configured some of what you asked, don’t reset anything. The likely issue is the complex task hung, hit model/tool failure, or failed sending the final reply. Paste the error lines from logs, with phone numbers/tokens redacted.

ionic dew Apr 30, 2026, 1:12 AM

#

Now it has come online briefly looks like it is talking but fails to say anything

fringe spruceBOT Apr 30, 2026, 1:13 AM

#

That “comes online / looks like typing / then says nothing” usually means:

WhatsApp received the message ✅
OpenClaw started processing ✅
then either the agent run failed or the final WhatsApp send failed ❌

Run this:

openclaw status --deep
openclaw channels status --probe
openclaw models status

Then open logs:

#

openclaw logs --follow

While logs are open, send one simple WhatsApp message like:

test

Look for the first error after that.

If it’s been stuck for an hour, also try:

/stop

#

in WhatsApp. If that doesn’t work:

openclaw gateway restart

Then test again with a simple message.

Paste the error lines from openclaw logs --follow — redact phone numbers/tokens. The log will tell us whether it’s model failure, stuck run, or WhatsApp send failure.

ionic dew Apr 30, 2026, 1:17 AM

#

How do I set up a second agent

twin matrixBOT Apr 30, 2026, 1:17 AM

#

fringe spruceBOT Apr 30, 2026, 1:17 AM

#

Yep — set up a second agent like this.

1. Pick the cheap/customer model ref first

openclaw models list

Copy the exact model name, like:

provider/cheap-model

2. Add a customer agent

openclaw agents add customer --workspace ~/.openclaw/workspace-customer --model provider/cheap-model --bind discord --non-interactive

#

Replace provider/cheap-model with your real cheap model.

3. Make sure WhatsApp stays on main

openclaw agents bind --agent main --bind whatsapp

4. Check routing

openclaw agents list --bindings

You want:

Discord -> customer

#

WhatsApp -> main

5. Restart

openclaw config validate
openclaw gateway restart

6. Check models per agent

openclaw models status --agent customer
openclaw models status --agent main

For the customer agent, also edit:

#

nano ~/.openclaw/workspace-customer/AGENTS.md

and tell it: answer customer/DJ booking questions, don’t code, don’t run scripts, keep replies short.

Docs: https://docs.openclaw.ai/cli/agents

ionic dew Apr 30, 2026, 1:18 AM

#

Can I do this via WhatsApp

fringe spruceBOT Apr 30, 2026, 1:18 AM

#

Sort of, but I wouldn’t recommend doing the setup itself via WhatsApp.

Adding agents/bindings is config/admin work, so do it in terminal where you can validate and restart safely.

Use terminal for this:

openclaw agents add customer --workspace ~/.openclaw/workspace-customer --model provider/cheap-model --bind discord --non-interactive
openclaw agents bind --agent main --bind whatsapp
openclaw config validate
openclaw gateway restart

Then you can use the agents from WhatsApp/Discord after that.

You can ask your bot in WhatsApp to help guide you, but I’d avoid letting it modify config while your WhatsApp channel has already been flaky. Terminal is safer.

ionic dew Apr 30, 2026, 9:34 AM

#

WhatsApp not responding is there a way the agent can fix and debug itself so I can make things stable

#I have my openclaw back up and running but the bot has not restored to its previous state

Simple setup: one bot, switch models when needed

Better cost-control setup: separate agents

Key idea

Option A — keep one agent, switch manually

Option B — add a second agent