#Discord error

1 messages · Page 1 of 1 (latest)

severe saffronBOT
#

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

tropic basinBOT
#

That message is the generic “agent run failed” fallback, so the useful bit will be in the gateway logs just before it sent that reply.

Can you paste:

openclaw --version
openclaw status --all
openclaw models status
openclaw logs --tail 200

Also try /new once in that Discord thread/DM. If /new fixes it, it was likely stale/corrupt session history. If it comes back immediately, the logs should show the real cause: auth refresh, missing provider auth, model/backend error, or Discord/channel delivery trouble.

Docs for the triage flow: https://docs.openclaw.ai/help/faq#first-60-seconds-if-something-is-broken and Discord/channel checks: https://docs.openclaw.ai/channels/troubleshooting#discord

#

Good, that narrows it: Discord transport and basic model replies are working. The failure is probably inside the agent/tool loop, not the channel itself.

Please run this, then trigger one failing tool-call request and paste the new log lines around the failure:

openclaw --version
openclaw models status
openclaw logs --follow

Also share which provider/model you’re using. Tool-call-only failures are commonly one of these:

  • the selected model/backend can chat but does not handle OpenClaw tool schemas reliably
  • local/OpenAI-compatible backend rejects structured tool payloads
  • a specific tool is denied/misconfigured and the beta is collapsing it into the generic error
  • provider auth refresh fails when the run enters the tool loop
#

That lines up with a model/tool compatibility failure. Plain chat working but tool turns failing means OpenClaw reaches OpenRouter, then the run breaks once tool schemas/results enter the loop.

Two next checks:

openclaw logs --tail 200

Trigger one failing tool request right before that, then look for the first real error above the generic Discord message. If it mentions tool schema, invalid request body, unsupported tool calls, structured content, or provider 4xx/5xx, that confirms it.

For a quick A/B test, switch the session to a stronger tool-capable model you have configured or one recommended in #models, then run the exact same tool request. If the other model works, DeepSeek v4 flash is the problem for tool use, not Discord.

Docs note the same failure shape here: https://docs.openclaw.ai/gateway/troubleshooting#local-openai-compatible-backend-passes-direct-probes-but-agent-runs-fail. Different backend, same pattern: chat can pass while agent/tool payloads fail.

#

Yep, that log confirms it: the failure is upstream from the model provider path, not Discord.

Key bits:

provider=openrouter
model=deepseek/deepseek-v4-flash
rawErrorPreview="500 Internal Server Error"
fallbackConfigured=false

So OpenClaw asked OpenRouter/DeepSeek to run the tool turn, got HTTP 500, and had no fallback model configured, so it surfaced the generic error.

Practical fix: switch your default away from openrouter/deepseek/deepseek-v4-flash for tool-heavy agent use, or configure at least one fallback model that handles tool calls well. You can use /model ... for a session-level test, or update agents.defaults.model once you’ve found one that works.

Docs: https://docs.openclaw.ai/concepts/models and fallback/config reference: https://docs.openclaw.ai/gateway/configuration-reference#models

#

If DeepSeek v4 flash is supposed to support this through OpenRouter, this is worth reporting with that log snippet to GitHub: https://github.com/openclaw/openclaw/issues. The important detail is “tool turns fail with OpenRouter DeepSeek v4 flash HTTP 500; plain chat works.”

#

Same 24-ish second failure across two OpenRouter models makes this look broader than one model. The heartbeat line is useful too: that is not Discord-specific; the embedded agent run itself is failing when it needs tool-capable behavior.

What I’d do next:

openclaw status --all
openclaw doctor
openclaw channels status --probe

Then A/B test outside Discord with the same model path:

openclaw infer model run --model openrouter/xiaomi/mimo-v2.5 --prompt "Say hello" --json

If plain infer works but Discord/tool/heartbeat runs fail, it’s almost certainly the OpenRouter model route choking on the agent/tool payload, not your Discord setup.