# Question about ollama+vision model support **URL:** \ **Category:** Support\ **Created:** [January 13, 2026, 12:43am UTC](https://forum.letta.com/t/question-about-ollama-vision-model-support/159 "2026-01-13T00:43:53Z")\ **Posts on this page:** 5\ **Page:** 1
**Author:** ![syd](https://yyz1.discourse-cdn.com/flex007/user_avatar/forum.letta.com/syd/32/118_2.png) [@syd](https://forum.letta.com/u/syd)\ **Post date:** [January 13, 2026, 12:43am UTC](https://forum.letta.com/t/question-about-ollama-vision-model-support/159/1 "2026-01-13T00:43:53Z")
I am trying to use Letta with a local Ollama backend and a vision-capable model (e.g. qwen2.5-vl, llama3.2-vision). However, Letta’s /v1/models endpoint only lists text-only Ollama models, and attempting to create an agent with an Ollama vision model fails Is it currently impossible to use Ollama + vision/multimodal models with Letta? Also, I was previously in the Letta Discord server and asked similar questions, but the server suddenly disappeared from my server list and I cannot rejoin via invite. If I was blocked, could you please let me know the reason? Thanks! ---
**Author:** ![cameron](https://yyz1.discourse-cdn.com/flex007/user_avatar/forum.letta.com/cameron/32/99_2.png) [@cameron](https://forum.letta.com/u/cameron)\ **Post date:** [January 15, 2026, 1:13am UTC](https://forum.letta.com/t/question-about-ollama-vision-model-support/159/2 "2026-01-15T01:13:19Z")
Unclear why you were blocked – can you send me your username? Regarding vision models, I was under the assumption that they worked. @ezra do you have any information here? ---
**Author:** ![ezra](https://yyz1.discourse-cdn.com/flex007/user_avatar/forum.letta.com/ezra/32/94_2.png) [@ezra](https://forum.letta.com/u/ezra)\ **Post date:** [January 15, 2026, 1:15am UTC](https://forum.letta.com/t/question-about-ollama-vision-model-support/159/3 "2026-01-15T01:15:09Z")
Here’s what I know about Ollama + vision: **Ollama has vision-capable models:** - `llama3.2-vision` - `llava` - `bakllava` - `moondream` **Letta multimodal support:** Per [docs.letta.com/multimodal](https://docs.letta.com/multimodal), image inputs work with compatible models. The docs explicitly list OpenAI (GPT-4o, etc.), Anthropic (Claude), and Google (Gemini) as supported. **The uncertainty:** I don’t see Ollama vision models explicitly listed in the multimodal docs. This could mean: 1. It works but isn’t documented 2. There are integration gaps with how Ollama handles image payloads 3. It depends on the specific model and how it expects image data **What would help diagnose:** - Which Ollama vision model are you trying to use? - What error or behavior are you seeing? - Are images being sent via URL or base64? If someone has tested Ollama + vision successfully (or hit specific errors), that would help clarify the current state. I don’t want to claim it works without verification. ---
**Author:** ![syd](https://yyz1.discourse-cdn.com/flex007/user_avatar/forum.letta.com/syd/32/118_2.png) [@syd](https://forum.letta.com/u/syd)\ **Post date:** [January 15, 2026, 7:10am UTC](https://forum.letta.com/t/question-about-ollama-vision-model-support/159/4 "2026-01-15T07:10:32Z")
Thank you for the response! My username is “sydsim” (display name is “시드”) About vision models, I asked the same question to ezra bot in letta discord server (it was the only chat I made except saying Hello) I have checked that qwen2.5 model is working in my local letta server with ollama, but when I try to use vision model (qwen2.5vl, llama3.2-vision), they do not appear in model list in letta dashboard and /v1/models endpoint, even though I can find them in ollama and it works fine when i try them independently in ollama. I also tried to create agent with Python sdk by explicitly providing the model name, it returned error like NOT\_FOUND: Handle ollama/qwen2.5vl:latest not found, must be one of [‘ollama/qwen2.5:latest’] ---
**Author:** ![cameron](https://yyz1.discourse-cdn.com/flex007/user_avatar/forum.letta.com/cameron/32/99_2.png) [@cameron](https://forum.letta.com/u/cameron)\ **Post date:** [January 20, 2026, 8:11pm UTC](https://forum.letta.com/t/question-about-ollama-vision-model-support/159/5 "2026-01-20T20:11:49Z")
> [@syd](#): > > sydsim You’ve been unbanned, apologies! TL;DR: This is an upstream Ollama limitation, not a Letta bug. Ollama doesn’t currently support tool calling for vision models like llama3.2-vision and qwen2.5-vl. Why vision models don’t appear: - Letta filters Ollama models to only show those with “tools” in their capabilities (via Ollama’s /api/show endpoint). This is intentional - Letta agents require tool calling to function (for memory operations, sending messages, etc.). - When you run ollama show llama3.2-vision, you’ll see the capabilities array likely only contains [“completion”, “vision”] but not “tools”. The upstream issue: - There’s an open GitHub issue tracking this: ollama/ollama#8345 - llama3.2-vision does not support tools ([llama3.2-vision does not support tools · Issue #8345 · ollama/ollama · GitHub](https://github.com/ollama/ollama/issues/8345)) Interestingly, Meta’s documentation says llama3.2-vision should support tool calling, but Ollama hasn’t implemented it yet for vision models. Once Ollama adds tool support for these models, they should automatically appear in Letta. Workaround: Unfortunately there isn’t a good workaround right now - without tool calling, the model can’t perform core agent functions like updating memory. You could use a non-vision Ollama model for the agent and handle image processing separately, but that’s not ideal. I’d recommend following the Ollama issue above for updates on when vision + tools will be supported.