#400 item error

1 messages · Page 1 of 1 (latest)

haughty scarab
#

Full error message:
400 Item 'rs_007e075d80b79184006994f6ca6da881948698694c9577ccd7' of type 'reasoning' was provided without its required following item.
model: gpt5-1-codex
openclaw version 2026.02.15

troubleshooting steps tried:
changing model to GPT-5-mini. GPT-5-1, GPT-5-2, GPT-5 - issue persisted
tried updating using installer command - no new version found

Thanks in advance

subtle muskBOT
#

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

summer prairieBOT
#

That 400 is coming from OpenAI’s Responses API, not your config. It happens when the session history OpenClaw replays contains an orphaned reasoning item (a reasoning block that isn’t followed by the required next item, e.g. text/tool call). Once that’s in the transcript, switching models won’t help until you clear/reset the session.

Fast workaround (usually fixes immediately)

In the same Discord thread / chat where it’s failing, send as a standalone message:

/new

(or /reset)

That starts a fresh session and drops the problematic history. (Docs mention this general “stale/corrupted session” fix too: https://docs.openclaw.ai/help/faq)

Why it often starts happening after 2–3 turns

This is commonly triggered if a turn gets aborted mid-generation (e.g. /stop, UI Stop button, disconnect, etc.) while using GPT-5* “reasoning” models, leaving a dangling reasoning block in the stored transcript.

Longer-term fix

#

This is a known bug in 2026.2.15 with OpenAI reasoning models + session replay. Updating to a newer build should fix it (there’s active work around “dropping orphaned OpenAI reasoning blocks”).

If you tell me these 2 things, I can point to the exact best path

  1. Install method: npm global, git, or macOS app?
  2. Did you use /stop (or the UI Stop) shortly before it started failing?

If /new fixes it, please right-click the solution message → Apps → ✅ Mark Solution so Answer Overflow indexes it for the next person.

haughty scarab
#

my install method was through powershell using iwr -useb https://openclaw.ai/install.ps1 | iex command

summer prairieBOT
#

Yep — that’s the Windows install.ps1 path.

For this specific 400 error, your install method isn’t the root cause; it’s almost always a bad session history (an orphaned OpenAI “reasoning” item). The quickest fix is:

  1. In the same Discord thread where it fails, send this as a standalone message:
/new

(or /reset)

That clears the session transcript for that thread, which is what’s poisoning follow-up turns.

If you want to try an update on Windows (to pick up the fix once it’s published)

Run the installer targeting the beta dist-tag:

# 
& ([scriptblock]::Create((iwr -useb https://openclaw.ai/install.ps1))) -Tag beta -NoOnboard

Then check:

openclaw --version

If -Tag beta still installs the same version, it just means beta == latest right now (so there’s nothing newer published yet).

Docs for the PS flags: https://docs.openclaw.ai/install/installer

If /new fixes it, please reply here and (once confirmed) mark the solution so it’s indexed for others (right-click → Apps → ✅ Mark Solution).

haughty scarab
#

/new and /reset does not fix the issue

vestal tulip
#

My Telegram keeps responsing with a 400 error, Ampere.sh chats dont response. have restated my instance multiple times via settings

summer prairieBOT
#

That sounds like the same underlying class of issue as the “reasoning required following item” 400s people are hitting with OpenAI/Responses models: a bad session transcript (usually after an aborted turn) poisons the chat history, and then every follow-up fails.

1) Quick fix for Telegram (do this first)

In the Telegram chat with the bot, send as a standalone message:

/new

(or /reset)

If it immediately works again, you’ve confirmed it’s session-history corruption (not your Telegram connection).

2) Check whether Ampere “doesn’t respond” vs “responds with an error”

Ampere/WebChat often just surfaces whatever the gateway run returned. If Telegram is getting 400s, Ampere may be failing the same way but not showing it clearly.

While sending a message from Ampere, run:

# 
openclaw logs --follow

…and see if you get a line like 400 Item 'rs_...' of type 'reasoning'... (or some other provider error).

3) Paste these 3 things here (they determine the next step)

  1. openclaw --version
  2. The exact 400 error text (the whole line, including the model name)
  3. openclaw channels status --probe (so we can see Telegram + WebChat health)

4) If you’re on 2026.2.15 and using OpenAI reasoning models

There are known issues around replaying OpenAI “reasoning” items in history. Workarounds until you can update:

  • Use /new whenever it gets stuck
  • Avoid /stop / aborting mid-response on those models (that’s a common trigger)
  • Temporarily switch that agent to a non-OpenAI model/provider (if that’s an option in your setup)
#

Once you paste the 3 outputs above, I’ll tell you whether this is the transcript/orphaned-reasoning case or a channel connectivity/routing issue.

vestal tulip
#

/new on Telegram got a response New session started. model: anthropic/claude-opus-4-6

#

follwed by the same 400 Something went wrong on our end message

#

openclaw logs --follow and openclaw --version on a new chat on Ampere.sh isnot giving back any response

summer prairieBOT
#

Those openclaw ... lines are terminal/PowerShell commands. If you typed them into an Ampere.sh chat message, nothing will happen (the bot won’t magically execute shell commands from chat text—especially if it’s currently failing to run/reply).

What I need is: run them on the machine where your OpenClaw Gateway is installed/running (PowerShell/Terminal), then paste the output here.

On Windows, open PowerShell on the gateway machine and run:

openclaw --version
openclaw status
openclaw gateway status
openclaw channels status --probe
openclaw logs --tail 200

Then paste the outputs (redact tokens/secrets).

Also: in Telegram, try sending /new as a standalone message to the bot once—if Telegram immediately starts working, it’s the orphaned “reasoning item” session-history issue and we can focus on update/workaround.

vestal tulip
summer prairieBOT
#

Ampere.sh is just a web UI client. It doesn’t run OpenClaw by itself — it connects to an OpenClaw Gateway that’s running somewhere.

So the question becomes: where is your Gateway running?

Case A) Gateway is on your Windows machine

Open Windows PowerShell (Start menu → type “PowerShell”) and run:

openclaw --version
openclaw status
openclaw gateway status
openclaw logs --tail 200

Case B) Gateway is on a Linux VPS (common with Ampere.sh / cloud)

You need a shell on that server:

Option 1: SSH from Windows PowerShell

# 
ssh <user>@<server-ip>

Then on the server:

openclaw --version
openclaw status
openclaw gateway status
openclaw logs --tail 200

Option 2: Ampere.sh “Console / Terminal”
Most VPS panels have a browser terminal. Open that, then run the same openclaw ... commands.

Case C) You’re using a fully hosted OpenClaw (no server access)

Then you can’t run openclaw CLI commands yourself — you’d need the host/provider’s logs/console.

#

Quick way to figure out which case you’re in

In Ampere.sh, what Gateway URL did you enter (or does it show) when you connected?

  • If it’s localhost / 127.0.0.1 → Case A
  • If it’s an IP/domain → Case B/C

Paste the gateway URL (redact token), and tell me whether you have an SSH login/IP for the machine. I’ll tell you exactly what to click/run next.

vestal tulip
summer prairieBOT
#

If it’s a fully hosted OpenClaw (Ampere manages the server), then you can’t run PowerShell/CLI on it — totally expected.

The good news: for this specific 400 “reasoning item missing” class of errors, you can usually fix it without server access by resetting the session (this is a gateway command, not a model call).

Do this from the chats themselves

  1. Telegram: send this as a standalone message (not a reply)
/new
  1. Ampere/Web chat: in the Ampere chat box, send:
/new

Then try a normal message again.

#

If /new doesn’t help

Your host needs to do one of these on the backend:

  • Delete/reset the stuck session (the transcript has an orphaned OpenAI “reasoning” item)
  • Update the hosted OpenClaw version (this is a known issue around 2026.2.15 with OpenAI Responses-style reasoning history replay)

Since you don’t have server access, the next thing I need from you is:

  • the exact 400 error text Telegram shows (paste it), and
  • if Ampere shows any error banner/toast, paste that too,
  • plus whatever OpenClaw version the Ampere UI shows (Status/About page, if available).

With that, I can tell you if it’s definitely the orphaned-reasoning transcript bug vs a connectivity/auth issue in the hosted setup.

vestal tulip