#i'm stuck with qwen2.5:14b

1 messages · Page 1 of 1 (latest)

rugged sierraBOT Feb 13, 2026, 1:39 AM

First: that Discord bot token is burned (it’s basically a password). Please delete the message and rotate/regenerate the token in the Discord Developer Portal now, then update your config/env var and restart the gateway. Don’t reuse the old one.

On the actual problem: the reply you’re seeing (“we can use the tts tool… / message function…”) is a classic symptom of a local Ollama model (often Qwen) producing “tool-y” garbage / meta text instead of behaving like a normal chat agent. In OpenClaw this is most commonly caused by streaming being enabled with Ollama models (there’s an upstream SDK formatting issue).

Fix (if you’re running Qwen via Ollama)

Make sure you’re on a recent OpenClaw version (streaming for Ollama is disabled by default in current versions).
Ensure you did not set streaming: true for your Ollama model. If you did, remove it, or explicitly set:

{
  agents: {
    defaults: {
      models: {
        "ollama/qwen2.5:14b": { streaming: false }
      }
    }

}
}

Restart the gateway.

Docs (this exact symptom is called out here):
https://docs.openclaw.ai/providers/ollama#corrupted-responses-or-tool-names-in-output

Also worth knowing

Even when it’s not “corrupted”, Qwen 14B is often shaky with tool-centric agent prompts. If you still get weird meta-responses after fixing streaming, try swapping the agent to a more tool-reliable model (ask in #models for current best picks):
https://docs.openclaw.ai/concepts/models

Discord config reference (mention gating, groupPolicy, etc.):
https://docs.openclaw.ai/channels/discord

If you paste more config, please redact tokens (or replace with "TOKEN_REDACTED"). Also tell me:

openclaw --version
whether you’re using Ollama (and if you manually enabled streaming anywhere)

cunning kindle Feb 13, 2026, 1:45 AM

{
"meta": {
"lastTouchedVersion": "2026.2.9",
"lastTouchedAt": "2026-02-11T21:31:23.830Z"
},
"wizard": {
"lastRunAt": "2026-02-11T21:31:23.824Z",
"lastRunVersion": "2026.2.9",
"lastRunCommand": "configure",
"lastRunMode": "local"
},
"models": {
"providers": {
"ollama": {
"baseUrl": "http://127.0.0.1:11434/v1",
"apiKey": "ollama-local",
"api": "openai-completions",
"models": [
{
"id": "qwen2.5:14b",
"name": "Qwen 2.5 14B",
"reasoning": false,
"input": [
"text",
"image"
],
"cost": {
"input": 0,
"output": 0,
"cacheRead": 0,
"cacheWrite": 0
},
"contextWindow": 65536,
"maxTokens": 16384
},
{
"id": "qwen2.5-coder:3b",
"name": "Qwen 2.5 Coder 3B",
"reasoning": false,
"input": [
"text"
],
"cost": {
"input": 0,
"output": 0,
"cacheRead": 0,
"cacheWrite": 0

},
"contextWindow": 65536,
"maxTokens": 16384
}
]
}
}
},
"agents": {
"defaults": {
"model": {
"primary": "ollama/qwen2.5:14b"
},
"maxConcurrent": 1,
"subagents": {
"maxConcurrent": 1
}
}
},
"commands": {
"native": "auto",
"nativeSkills": "auto"
},
"gateway": {
"auth": {
"mode": "token",
"token": "cada95--------------------"
},
"remote": {
"token": "cada9----------------------"
},
"mode": "local"
},
"messages": {
"ackReactionScope": "group-mentions"
},
"plugins": {
"entries": {
"discord": {
"enabled": true
}
}
},
"channels": {
"discord": {
"enabled": true,
"token": "MT------------------",
"groupPolicy": "open"
}
}
}

and my discord bot error: It looks like you're simulating a Discord message scenario. If your intention is to send a message in the Discord channel with ID 147----------------, we can use the tts tool if it's an audio message or the message function if it's a text chat message.

However, from your prompt, it seems that you're simulating sending a Discord message without actually needing to perform any actions with the provided tools. If you want me to send a message in Discord using one of the available functions, please specify whether you'd like this to be an audio or text message and provide additional context or parameters.

If your goal is merely documentation or analysis (such as saving the interaction details), we can proceed without invoking any function calls. Please clarify your objective so I can assist effectively!

rugged sierraBOT Feb 13, 2026, 1:51 AM

cunning kindle }, "contextWindow": 65536, "maxTokens": 16384 ...

The snippet you pasted explains why you’re seeing that weird “I can use the tts tool / message function…” style response: you’re running a local Ollama model (qwen2.5:14b) and it’s not behaving well with OpenClaw’s tool/agent prompt.

Two concrete things to do:

0) Security (seriously)

You posted a real Discord bot token earlier in this thread. Please regenerate/rotate it in the Discord Developer Portal and restart the gateway. Even if it’s deleted now, assume it was scraped.

1) Force Ollama streaming OFF for that model

If you configured Ollama via models.providers.ollama (your earlier snippet with contextWindow / maxTokens looks like that), you likely didn’t inherit the “streaming disabled” default that auto-discovery applies.

Add an explicit model entry override:

{
  agents: {
    defaults: {

      models: {
        "ollama/qwen2.5:14b": {
          streaming: false
        }
      }
    }
  }
}

Then restart the gateway.

Why: there’s a known upstream issue where Ollama + streaming can cause “tool-ish / meta” output instead of normal agent behavior. Docs: https://docs.openclaw.ai/providers/ollama#corrupted-responses-or-tool-names-in-output

2) Reality check: qwen2.5:14b is often just not tool-reliable

Even with streaming off, smaller local models frequently don’t do consistent tool calling and will “explain tools” instead of using them. If streaming-off doesn’t fix it, try temporarily switching your primary model to a known tool-capable Ollama model to confirm it’s model-related (not Discord config).

If you paste one more thing, I can be precise: your full models: { ... } block for Ollama (with tokens removed), plus openclaw --version.

That text is not a Discord API/bot “error” — it’s your model (qwen2.5:14b) replying with meta nonsense instead of acting like the agent. In other words: Discord is delivering messages fine; the LLM is just not following the agent/tooling prompt.

1) Make sure Ollama streaming is OFF (common cause)

In current OpenClaw, Ollama models default to non-streaming to avoid a known upstream issue — but if you’re on an older version or you overrode it, you can force it off explicitly:

{
  agents: {
    defaults: {
      models: {
        "ollama/qwen2.5:14b": {
          streaming: false
        }
      }
    }
  }

Restart gateway after changing config.

Ref: https://docs.openclaw.ai/providers/ollama#corrupted-responses-or-tool-names-in-output

2) Quick workaround: hide tools from Qwen (so it stops talking about them)

If your immediate goal is “just chat in Discord”, you can restrict tools for that model:

{
  tools: {
    byProvider: {
      "ollama/qwen2.5:14b": { profile: "minimal" }