help | Friends of the Crustacean 🦞🤝 | Page 1

ruby hound Apr 18, 2026, 5:38 PM

#

17:33:06 [agent/embedded] incomplete turn detected: runId=954fe449-bf4d-4fa0-8a14-8e895bc7ec53 sessionId=46d1f0a8-201c-4462-b159-b0de5e1ef907 stopReason=stop payloads=0 — surfacing error to user
17:37:34 [agent/embedded] incomplete turn detected: runId=9f2dfc16-8c4c-442b-93cc-c74b8269396e sessionId=46d1f0a8-201c-4462-b159-b0de5e1ef907 stopReason=stop payloads=0 — surfacing error to user

wary daggerBOT Apr 18, 2026, 5:38 PM

#

wheat zenithBOT Apr 18, 2026, 5:38 PM

#

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

wary daggerBOT Apr 18, 2026, 5:38 PM

#

frosty dewBOT Apr 18, 2026, 5:40 PM

#

That line is a symptom, not the root cause.

stopReason=stop + payloads=0 means the embedded run finished without any user-visible reply. In practice that usually means one of these:

the model returned an empty response
it returned reasoning/internal-only output but no final answer
upstream hit a rate limit / overload / context issue and the turn collapsed before reply

A normal tool-call stall is usually stopReason=toolUse, so this one points more at an empty/hidden final turn.

Please paste these:

openclaw --version
openclaw status --all
openclaw models status

#

Then run:

openclaw logs --follow

and trigger it once more. In the logs, look for lines right before the incomplete turn detected one, especially:

reasoning-only assistant turn detected
empty response detected
rate limit
overloaded
context overflow

If you're using OpenAI / GPT-style models, there was a recent fix for exactly these reasoning-only / empty embedded turns, so if you're on an older build I'd upgrade first.

If you're using a local OpenAI-compatible backend (Ollama, LM Studio, llama.cpp, etc.), tell me the exact backend + model too, because those are a very common cause of empty-response turns.

#

Docs:

ruby hound Apr 18, 2026, 5:45 PM

#

Diagnosis (read-only)

Gateway connection details:
Gateway target: ws://127.0.0.1:18789
Source: local loopback
Config: /root/.openclaw/openclaw.json
Bind: loopback

✓ Config: /root/.openclaw/openclaw.json
✓ Secret diagnostics (0)
✓ Restart sentinel: none
! Port 18789
Port 18789 is already in use.

pid 9778 root: openclaw-gateway (127.0.0.1:18789)
Gateway already running locally. Stop it (openclaw gateway stop) or use a different port.
! Tailscale: off · unknown
✓ Skills: 7 eligible · 0 missing · /root/.openclaw/workspace
✓ Plugin compatibility (none)
✓ Channel issues (none)

Pasteable debug report. Auth tokens redacted.
Troubleshooting: https://docs.openclaw.ai/troubleshooting

◇

🦞 OpenClaw 2026.4.15 (041266a) — I'm the middleware between your ambition and your attention span.

Config : ~/.openclaw/openclaw.json
Agent dir : ~/.openclaw/agents/main/agent
Default : openrouter/deepseek/deepseek-v3.2
Fallbacks (0) : -
Image model : -
Image fallbacks (0): -
Aliases (1) : OpenRouter -> openrouter/auto
Configured models (2): openrouter/auto, openrouter/deepseek/deepseek-v3.2

Auth overview
Auth store : ~/.openclaw/agents/main/agent/auth-profiles.json
Shell env : off
Providers w/ OAuth/tokens (0): -

openrouter effective=profiles:~/.openclaw/agents/main/agent/auth-profiles.json | profiles=1 (oauth=0, token=0, api_key=1) | openrouter:default=sk-or-v1...4f10cb48

OAuth/token status

none
root@ubuntu:~#

frosty dewBOT Apr 18, 2026, 5:46 PM

#

Nice, this narrows it down a lot.

A few important takeaways from that report:

Gateway looks up/alive
2026.4.15 is new enough that the recent empty/reasoning-only turn recovery fixes are already in
Port 18789 is already in use just means your gateway is already running, not that it failed

So this no longer looks like a gateway/startup issue.
The most suspicious part is now the model path: openrouter/deepseek/deepseek-v3.2 with no fallbacks.

I’d test it in this order:

openclaw models status --probe
openclaw infer model run --model openrouter/deepseek/deepseek-v3.2 --prompt "Reply with exactly: OK" --json

#

Then also:

openclaw logs --follow

and trigger one failing chat turn. The lines right before incomplete turn detected matter most.

What I’d expect:

if infer model run also fails, it’s basically a provider/model issue
if infer model run works but chat still fails, the model is probably choking on the full agent prompt/tool context
if a fresh session works but the current one doesn’t, it may be session/context related, so try /new

Honestly, DeepSeek via OpenRouter would be my prime suspect here, not the gateway.

If you want the fastest unblock, I’d temporarily switch to a more tool-reliable model, or at least add a fallback, then compare behavior.

#

If you paste:

openclaw models status --probe
the output of that infer model run
~20 log lines before the failure

I can help pinpoint whether this is OpenRouter, the model, or agent/tool prompt pressure.

Docs:

ruby hound Apr 18, 2026, 5:49 PM

#

🦞 OpenClaw 2026.4.15 (041266a) — Finally, a use for that always-on Mac Mini under your desk.

│
◇
Config : ~/.openclaw/openclaw.json
Agent dir : ~/.openclaw/agents/main/agent
Default : openrouter/deepseek/deepseek-v3.2
Fallbacks (0) : -
Image model : -
Image fallbacks (0): -
Aliases (1) : OpenRouter -> openrouter/auto
Configured models (2): openrouter/auto, openrouter/deepseek/deepseek-v3.2

Auth overview
Auth store : ~/.openclaw/agents/main/agent/auth-profiles.json
Shell env : off
Providers w/ OAuth/tokens (0): -

openrouter effective=profiles:~/.openclaw/agents/main/agent/auth-profiles.json | profiles=1 (oauth=0, token=0, api_key=1) | openrouter:default=sk-or-v1...4f10cb48

OAuth/token status

none

Auth probes
┌───────────────────────────────────┬──────────────────────────────┬────────────┐
│ Model │ Profile │ Status │
├───────────────────────────────────┼──────────────────────────────┼────────────┤
│ openrouter/deepseek/deepseek-v3.2 │ openrouter:default (api_key) │ ok · 11.3s │
└───────────────────────────────────┴──────────────────────────────┴────────────┘
Probed 1 target in 17.8s
17:48:06 [agent/embedded] incomplete turn detected: runId=46d1f0a8-201c-4462-b159-b0de5e1ef907 sessionId=46d1f0a8-201c-4462-b159-b0de5e1ef907 stopReason=stop payloads=0 — surfacing error to user
{
"ok": true,
"capability": "model.run",
"transport": "local",
"provider": "openrouter",
"model": "deepseek/deepseek-v3.2",
"attempts": [],
"outputs": [
{
"text": "⚠️ Agent couldn't generate a response. Please try again.",
"mediaUrl": null
}
]
}
root@ubuntu:~#

frosty dewBOT Apr 18, 2026, 5:50 PM

#

Yep, this basically pins it down.

Your setup is authing and reaching the model fine:

models status --probe = ok
but a trivial infer model run on the same model still returns:
⚠️ Agent couldn't generate a response

So the failure is not:

gateway startup
Discord delivery
session history
tool calling

It’s lower-level than that: that model route is ending cleanly but not producing usable text.

#

So at this point I’d treat openrouter/deepseek/deepseek-v3.2 as the problem.

Fastest unblock

Temporarily switch off that model and test another one.

Since you already have openrouter/auto configured, easiest test is:

openclaw models set openrouter/auto
openclaw infer model run --model openrouter/auto --prompt "Reply with exactly: OK" --json

If that works, you’ve confirmed it’s specific to the DeepSeek V3.2 OpenRouter route, not OpenClaw itself.

What this means

#

models status --probe only proves auth + route reachability.
It does not prove the model will return a valid final answer for normal inference.

Your repro now is strong enough to say:

OpenClaw can reach/auth OpenRouter DeepSeek V3.2, but actual inference returns an empty/incomplete final turn.

What I’d do next

Use another primary model for now
Add a fallback so the bot doesn’t hard-fail on this route
If you want to help get it fixed, file a GitHub issue with exactly these two outputs:
- openclaw models status --probe
- openclaw infer model run --model openrouter/deepseek/deepseek-v3.2 --prompt "Reply with exactly: OK" --json

Issue tracker:
https://github.com/openclaw/openclaw/issues

#help

Fastest unblock

What this means

What I’d do next