Ollama

Ollama "model "llama3" not found, try pulling it first"

Last checked

The error

{"error":"model \"llama3\" not found, try pulling it first"}

Ollama embed failed: HTTP 404 Not Found {"error":"model \"all-MiniLM-L6-v2\" not found, try pulling it first"}

HTTP 404 from the Ollama API. The second line is how an OpenClaw memory plugin surfaced it for an embedding model. Current builds return model 'llama3' not found for chat and generate calls; same cause.

The Ollama server that received the request has no model with that exact name and tag. Usually the model is on your machine, but your agent is talking to a different Ollama (a Docker container, a system service with its own model folder, a remote host), or it asks for llama3 when you pulled llama3:8b. Run curl on /api/tags from where the agent runs, then pull the exact name on that server.

Why it happens

Ollama looks up models only in its own models directory, and a name with no tag means name:latest. Anything that makes the agent's server, folder or tag differ from the one you pulled into produces this 404.

  1. A different Ollama instance. Ollama in Docker keeps models in its own volume at /root/.ollama, so models pulled on the host are invisible to it, and the reverse. The same happens when the agent's base URL points at another machine.
  2. A different models folder. The Linux installer runs Ollama as the ollama user with models in /usr/share/ollama/.ollama/models, while running ollama serve by hand uses your home folder. OLLAMA_MODELS set for one and not the other splits them too.
  3. Tag mismatch. llama3 means llama3:latest. If you only pulled llama3:8b or a quantized tag, the bare name is not found. Names must match what ollama list shows.
  4. The embedding model was never pulled. Agents with memory or RAG call a separate embedding model, and that is often the one missing, as in the OpenClaw memory plugin report above.
  5. A typo or a provider prefix left in the model id, for example sending ollama/llama3 as the model name to the Ollama API.

The fix

  1. 1 From the machine or container where the agent runs, list what that server has: curl http://127.0.0.1:11434/api/tags (use the same host and port as the agent's base URL).
  2. 2 Compare the names to the model id in your agent config. Copy the exact name:tag from the list, including the tag.
  3. 3 If the model is missing there, pull it on that server. For the CLI against a specific server: OLLAMA_HOST=http://host:11434 ollama pull llama3:8b. For the official Docker container: docker exec -it ollama ollama pull llama3:8b.
  4. 4 If you use a custom OLLAMA_MODELS path on Linux, set it in the systemd service with systemctl edit ollama.service and give the ollama user access with sudo chown -R ollama:ollama on that directory, then restart.
  5. 5 Pull the embedding model your agent's memory or RAG setting names, not just the chat model.
  6. 6 Restart the agent so it re-reads the model list.
curl http://127.0.0.1:11434/api/tags

Which wording will you see?

Both strings come from Ollama's server code. Chat and generate requests on current builds check the model first and return model 'llama3' not found with single quotes. Requests that reach the scheduler without that check, such as the legacy embeddings endpoint, return model "llama3" not found, try pulling it first. Older Ollama builds and many agent logs show the second form. Either way it is an HTTP 404 and the fixes are the same.

Still failing?

  • Run ollama list and curl /api/tags side by side: if they differ, your CLI and your agent are talking to different servers.
  • Check the agent's base URL for a Docker-only hostname such as host.docker.internal or a container name that resolves to a different Ollama than you expect.
  • Look for a second Ollama process (ps aux | grep ollama, or Task Manager on Windows) that started first and holds port 11434.

Related errors

Full guideOpenClaw Local Model Not Working? Here's Why (And What Actually Fixes It)Every error, one pageOpenClaw + Ollama: What Works and What Doesn't (2026)

Hit a different error?

Paste any agent error and get the cause and fix in seconds.

Open the decoder

Frequently asked questions

ollama run works, so why does my agent get this error?

ollama run talks to whatever server OLLAMA_HOST points at, and pulls automatically if needed. Your agent may hit a different server, a different models folder, or ask for a name without the tag you pulled.

Do I need the :latest tag?

A name without a tag is treated as name:latest. If ollama list only shows llama3:8b, either use llama3:8b in the agent or pull llama3 as well.

I pulled the model on my host but Ollama runs in Docker. What now?

Pull it inside the container with docker exec, or point the agent at the host's Ollama instead. Two Ollama installs never share models unless they share the same models directory.

Stop firefighting agent errors

Decoding errors one at a time is the manual version of what BetterClaw automates. Run your agents on a no-code AI agent platform with managed models, retries and config validation built in.

Deploy Your Agent — Start FreeBook a Demo

Free plan available · Pro $49/mo · BYOK · 7-day money-back guarantee