#Sending images via telegram gives cryptic error loop

1 messages · Page 1 of 1 (latest)

ebon latch
#

When sending an image via telegram i get "LLM request failed: provider rejected the request schema or tool payload." returned even though i have an imagemodel set both models on ollama cloud. everey following message thereafter gets same error back session needs to be restarted.

topaz spindleBOT
#

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

hidden solsticeBOT
split depotBOT
#

Let me check the docs and source so I don’t guess here.

#

I found the exact error mapping in source. Now I’m checking how image routing works with imageModel and whether Ollama/OpenAI-compat vision is the real mismatch.

#

That error is annoyingly generic, but with Ollama the two most likely causes are:

  1. You’re using Ollama through /v1 or another OpenAI-compatible path, instead of the native Ollama API.
    For Ollama, OpenClaw wants api: "ollama" and a baseUrl with no /v1. The docs explicitly warn that /v1 breaks tool calling.
    Docs: https://docs.openclaw.ai/providers/ollama

  2. imageModel is not the same as “always use this for images.”
    OpenClaw only uses agents.defaults.imageModel when the active model cannot accept images. If your current session model is marked as image-capable, OpenClaw will send the image there first. If that model/provider then rejects the image payload, you get exactly this kind of schema error.
    Docs: https://docs.openclaw.ai/concepts/models

Also, the “every message after that keeps failing until restart” part sounds like the bad turn is sticking in session history. Try /new instead of restarting the whole session. If /new fixes it, that points to replay/session state getting poisoned by the failed multimodal turn.

Can you paste these so I can narrow it down?

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

Then reproduce once while tailing logs:

openclaw logs --follow

What I especially want to see is:

  • whether your Ollama provider is using api: "ollama" and no /v1
  • which model is actually active for that Telegram session
  • the raw 400/422 log line behind the generic “schema or tool payload” message

Helpful refs:

ebon latch
#

yeah i had image defined for glm5.1 remnant from experimentations removed it and kimi2.6 s set as image model but error persists

split depotBOT
#

Yep, then I’d check these next, in this order:

  1. Start a fresh session

    • use /new in that Telegram chat
    • if you changed defaults, the old session can still be stuck on stale replay/model state

  2. Check what model that Telegram session is actually using

    • run /model status in the Telegram chat
    • if it still shows the old GLM/session override, changing defaults won’t help that existing session

  3. If you defined Ollama models manually, make sure the Kimi image model entry itself is marked image-capable:

    input: ["text", "image"]

    Just setting agents.defaults.imageModel is not enough if the model catalog entry says text-only.

  4. Make sure Ollama is native mode, not /v1

#
  • api: "ollama"
    • base URL with no /v1

If it still fails after /new, please paste:

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

and then send one test image so we can see the raw error behind the generic message.

My guess right now is either:

  • the Telegram session is still pinned to stale model/session state, or
  • the Kimi model entry is still not being treated as input: ["text","image"] by the catalog.
ebon latch
#

yes just a malconfiguration from experimenting earlier thx