New session error | Friends of the Crustacean 🦞🤝 | Page 1

inland spear Apr 29, 2026, 11:51 PM

#

Every /new while using OpenAI/gpt returns: “One of "input" or "previous_response_id"or 'prompt'or 'conversation_id' must be provided.”

dark wyvernBOT Apr 29, 2026, 11:51 PM

#

To help others find answers, you can mark your question as solved via Right click solution message -> Apps -> ✅ Mark Solution

vagrant summitBOT Apr 29, 2026, 11:53 PM

#

That looks like OpenAI Responses getting an empty request after /new — it needs input or a state pointer like previous_response_id/conversation state. OpenAI’s own examples create Responses with input, and the API supports previous_response_id for multi-turn state. (platform.openai.com)

I’d check two things first:

openclaw --version
openclaw config get agents.defaults.model
openclaw config get agents.defaults.models

If your model is openai/gpt-*, try disabling the OpenAI WebSocket warm-up / forcing SSE for that model:

openclaw config set agents.defaults.models '{"openai/gpt-5.5":{"params":{"openaiWsWarmup":false,"transport":"sse"}}}' --strict-json --merge
openclaw gateway restart

#

Replace openai/gpt-5.5 with your exact model ref.

If you’re actually on openai-codex/gpt-*, current main has an unreleased fix for empty Codex Responses input (input: [], issue #73820), so updating to the next release/main should fix it. If the workaround doesn’t help, paste openclaw status --all and the relevant log tail. Docs: https://docs.openclaw.ai/providers/openai

inland spear Apr 30, 2026, 12:12 AM

#

aiuser@Aouns-iMac ~ % openclaw --version
OpenClaw 2026.4.26 (be8c246)
aiuser@Aouns-iMac ~ % openclaw config get agents.defaults.model

🦞 OpenClaw 2026.4.26 (be8c246)
Your second brain, except this one actually remembers where you left things.

{
"primary": "openai-codex/gpt-5.5",
"fallbacks": [
"openai-codex/gpt-5.5",
"openai-codex/gpt-5.4",
"openai-codex/gpt-5.4-mini",
"openai-codex/gpt-5.3-codex",
"nvidia/deepseek-ai/deepseek-v3.1",
"openrouter/nvidia/nemotron-3-super-120b-a12b:free",
"openrouter/openrouter/free",
"openrouter/minimax/minimax-m2.5:free",
"openrouter/stepfun/step-3.5-flash:free"
]
}
aiuser@Aouns-iMac ~ % openclaw config get agents.defaults.models

🦞 OpenClaw 2026.4.26 (be8c246)
Your terminal just grew claws—type something and let the bot pinch the busywork.

{
"nvidia/deepseek-ai/deepseek-v3.1": {},
"openrouter/minimax/minimax-m2.5:free": {},
"openrouter/nvidia/nemotron-3-super-120b-a12b:free": {},
"openrouter/stepfun/step-3.5-flash:free": {},
"openai-codex/gpt-5.4": {},
"openai-codex/gpt-5.4-pro": {},
"openai-codex/gpt-5.4-mini": {},
"openai-codex/gpt-5.3-codex": {},
"openai-codex/gpt-5.2": {},
"openai-codex/gpt-5.1": {},
"openai-codex/gpt-5.1-codex-max": {},
"openai-codex/gpt-5.1-codex-mini": {},
"openai-codex/gpt-5.2-codex": {},
"openai-codex/gpt-5.3-codex-spark": {},
"openrouter/openrouter/free": {},
"openrouter/minimax/minimax-m2.7": {},
"openrouter/deepseek/deepseek-v3.2": {},
"openrouter/moonshotai/kimi-k2.5": {},
"openrouter/openrouter/auto": {},
"openai-codex/gpt-5.5": {},
"openai-codex/gpt-5.5-pro": {},
"google/gemini-3.1-pro-preview": {
"alias": "gemini"
},
"google/gemini-3.1-flash-lite-preview": {
"alias": "gemini-flash-lite"
}
}
aiuser@Aouns-iMac ~ %

vagrant summitBOT Apr 30, 2026, 12:12 AM

#

Yep — you’re on the affected version/path. Your config is using openai-codex/gpt-5.5, and this exact empty Codex Responses request was fixed after 2026.4.26.

Do this first:

openclaw update
openclaw --version

You should land on 2026.4.27 or newer. Then retry /new.

If you need a temporary workaround before updating, switch the primary to a non-Codex fallback or another provider. I wouldn’t spend more time tweaking this config on 2026.4.26 — it’s a known Codex Responses bug.

Docs: https://docs.openclaw.ai/install/updating

inland spear Apr 30, 2026, 12:19 AM

#

Ok I updated but what are these warning about;

Updating plugins...
No plugin updates needed.
Completion cache update failed: Error: spawnSync /opt/homebrew/Cellar/node/25.8.1_1/bin/node ETIMEDOUT

Restarting service...
Gateway did not become healthy after restart.
Gateway version mismatch: expected 2026.4.27, running gateway reported unavailable.
Service runtime: status=running, state=active, pid=97411
Gateway port 18789 status: free.
Restart log: /Users/aiuser/.openclaw/logs/gateway-restart.log
Run openclaw gateway status --deep for details.

Gateway is up and running btw.

vagrant summitBOT Apr 30, 2026, 12:19 AM

#

Those are mostly post-update housekeeping warnings, not necessarily fatal.

Completion cache update failed ... ETIMEDOUT
This is just the shell tab-completion cache generation timing out. It doesn’t affect the gateway or /new. If you care about CLI completions, run later:
```
OPENCLAW_COMPLETION_SKIP_PLUGIN_COMMANDS=1 openclaw completion --write-state
```
Gateway did not become healthy after restart + version ... unavailable
The updater restarted the macOS service, then its health probe couldn’t reach the gateway quickly enough. Since it says port 18789 was free, that usually means the service process existed but the gateway wasn’t listening yet during the check — not a port conflict.

If it’s up now, verify it’s actually the new binary:

openclaw --version
openclaw gateway status --deep

#

openclaw status --all

As long as those show 2026.4.27 and gateway reachable, you’re fine. If it still reports unavailable/mismatch, run:

openclaw gateway restart

One small flag: your Node path is Node 25.8.1. OpenClaw docs recommend Node 24 by default, with Node 22.14+ supported. If weird timeouts keep happening, I’d move back to Node 24 later. Docs: https://docs.openclaw.ai/install/node

inland spear Apr 30, 2026, 12:32 AM

#

Everything looks fine with the versions. But now after the update when I did /new it returned: “✅ Done.”

vagrant summitBOT Apr 30, 2026, 12:33 AM

#

Yep — that’s the expected success response for /new. It means the session reset completed cleanly.

The old error was happening while resetting/starting the fresh Codex Responses session; after the update, ✅ Done. means that path is fixed.

Try sending a normal message after /new. If the assistant replies normally, you’re good 🎉

If this fixed it, please right-click the solution message → Apps → ✅ Mark Solution so Answer Overflow indexes it for future users.

inland spear Apr 30, 2026, 12:35 AM

#

I never got that before. Usually after /new I get something like this: Morning Xxxx — Chris here ⚡
I’m online and in the right headspace. What do you want to tackle first?

vagrant summitBOT Apr 30, 2026, 12:35 AM

#

Ah, yep — that’s the distinction:

Bare /new now just resets the session and returns the command ack: ✅ Done.
It does not invoke the model, so you won’t get the “Morning … I’m online…” generated greeting from /new alone.

That matches the current docs: bare /new / /reset are acknowledged without calling the model. https://docs.openclaw.ai/start/openclaw

If you want the old style, do either:

/new
hi

or send a message with /new, e.g.:

#

/new say your startup greeting

If the text after /new isn’t a model name, OpenClaw treats it as the first message body.

inland spear Apr 30, 2026, 12:52 AM

#

Oh so that’s the new expected behavior?

vagrant summitBOT Apr 30, 2026, 12:53 AM

#

Yep — for a bare /new, ✅ Done. is the expected behavior now.

It resets the session without spending a model call. The greeting happens on the next normal message, or if you include text after /new.

So:

/new

→ reset only, ✅ Done.

/new what are we doing today?

→ reset, then model responds to that message.

inland spear Apr 30, 2026, 12:55 AM

#

Ok. What did you say about my node version? I should downgrade it? Whatever version I have got installed with OpenClaw.

vagrant summitBOT Apr 30, 2026, 12:55 AM

#

Not urgent — don’t downgrade just because of that warning if everything is working now.

What I meant: your log showed Node 25.8.1. OpenClaw requires Node 22.14+, but the recommended/default runtime is Node 24. Node 25 is newer, but less “boring/stable” for OpenClaw testing.

My recommendation:

If /new, gateway, and normal replies are working: leave it alone for now.
If you keep seeing weird timeout/update/gateway issues: switch to Node 24, then reinstall/rewrite the gateway service so launchd points at the right Node.

Docs: https://docs.openclaw.ai/install/node

inland spear Apr 30, 2026, 1:17 AM

#

Was my model config okay?

vagrant summitBOT Apr 30, 2026, 1:17 AM

#

Structurally, yes — your config looked OK.

What I saw:

agents.defaults.model.primary = openai-codex/gpt-5.5
That same model is present in agents.defaults.models, so it’s allowed/selectable.
Your fallback list entries also appeared to be present in the allowlist.
Having the primary repeated as the first fallback is harmless, just redundant.

The only caveat: model availability can change by account/provider, so validate with:

openclaw models status
openclaw models list --provider openai-codex

If those look healthy and normal messages work after /new, your model config is fine.

inland spear Apr 30, 2026, 1:54 AM

#

What’s the right way to add anthropic models to my config using claude-cli?

vagrant summitBOT Apr 30, 2026, 1:55 AM

#

Yes — the new preferred shape is:

model ref stays canonical: anthropic/<model-id>
execution backend is separate: agentRuntime.id = "claude-cli"
don’t use new claude-cli/... model refs unless you’re keeping legacy compatibility.

Flow:

claude --version
openclaw onboard
# choose Claude CLI
openclaw models list --provider anthropic

Pick the exact model id from that list, then:

#

openclaw config set agents.defaults.agentRuntime.id claude-cli
openclaw config set agents.defaults.models '{"anthropic/<model-id>":{}}' --strict-json --merge
openclaw config set agents.defaults.model.primary anthropic/<model-id>
openclaw gateway restart

If claude isn’t visible to the launchd gateway, set the absolute path too:

which claude
openclaw config set 'agents.defaults.cliBackends["claude-cli"].command' "/opt/homebrew/bin/claude"
openclaw gateway restart

Then start fresh with /new.

#

Docs: https://docs.openclaw.ai/providers/anthropic and https://docs.openclaw.ai/concepts/agent-runtimes

inland spear Apr 30, 2026, 2:00 AM

#

Should I be using onboard or configure on my existing setup?

vagrant summitBOT Apr 30, 2026, 2:01 AM

#

For an existing setup, use:

openclaw configure --section model

onboard is more of the first-run / broad setup wizard. It’s fine to rerun, but for “add Claude CLI / Anthropic to my current config” it’s overkill.

Important nuance: configure --section model merges provider/model choices into your existing allowlist and generally preserves your current primary. If you want Claude CLI to become the default too, set that explicitly after:

openclaw models set anthropic/<model-id>
openclaw config set agents.defaults.agentRuntime.id claude-cli
openclaw gateway restart

Use the exact model id from:

#

openclaw models list --provider anthropic

Docs: https://docs.openclaw.ai/cli/configure and https://docs.openclaw.ai/providers/anthropic

#New session error