How to Fix Ollama "Error: Pull Model Manifest" & Connection Refused on Local PC

Ollama's "Error: pull model manifest" usually means it couldn't retrieve or process the model's metadata from the registry. A "connection refused" error usually means the Ollama server isn't accepting connections at its expected address, commonly 127.0.0.1:11434. Start by checking whether Ollama is running, then verify the model name and test the local API before reinstalling anything.

Quick Diagnostic Table

Error or Symptom	Likely Cause	First Fix	Estimated Time
Connection refused at localhost:11434	Ollama server isn't running or listening	Restart Ollama	2–5 minutes
Pull model manifest error	Invalid model name, tag, or registry access problem	Check the model name and tag	2–10 minutes
Model downloads never start	Internet, DNS, proxy, or registry connection issue	Test internet and proxy settings	5–15 minutes
A web interface cannot connect to Ollama	Incorrect API URL or host configuration	Check the interface's Ollama endpoint	3–10 minutes
Pull starts but fails partway through	Interrupted connection, disk space, or download problem	Check storage and retry	5–20 minutes

These times are estimates, not guaranteed repair times. The cause depends on which error appears and whether you're using Ollama directly, through a terminal, or through another application.

What Is Actually Triggering Ollama Manifest and Connection Errors?

There are two problems that often get mixed together.

A manifest error happens when Ollama can't retrieve or process the metadata describing a model and its available files. This can happen because the model name or tag is incorrect, the requested model isn't available under that reference, or the registry can't be reached.

A connection-refused error is different. Your application tries to contact Ollama's local API, but nothing is accepting connections at the requested address and port.

By default, Ollama's local API commonly uses:

http://127.0.0.1:11434

That address points back to your own computer. It doesn't refer to an external server.

The usual causes include:

• Ollama isn't running: The desktop application or background server has stopped.
• The model reference is incorrect: A typo in the model name or tag prevents Ollama from finding the requested manifest.
• The registry is unreachable: DNS, internet connectivity, proxy settings, or a temporary service problem prevents the download.
• Port or host configuration problems: Ollama is listening on a different address, or another process is using the expected port.
• Insufficient storage: The model download can't complete because the destination drive is nearly full.
• An application is using the wrong endpoint: A local AI interface may be configured to connect to the wrong host or port.

Real-World Tip: Don't assume every manifest error requires a reinstall. There's no verified universal 80% success rate for one fix. First test http://localhost:11434 and run ollama list. These checks quickly show whether the local server responds and whether Ollama can access its model inventory.

Proven Fixes for Ollama "Error: Pull Model Manifest" & Connection Refused

1. Restart Ollama and Check the Local API

Start with the simplest fix. If you're using Ollama on Windows, look for its icon in the system tray near the clock. If it's running, exit the application from its tray menu, then open Ollama again from the Start menu.

If you run Ollama as a service on Linux, check its status:

sudo systemctl status ollama

If the service is stopped or needs restarting, run:

sudo systemctl restart ollama

On macOS, restart the Ollama application from Applications if you're using the desktop app.

Now open a browser and enter:

http://localhost:11434

When the local server is responding, a plain response such as Ollama is running is normal. This confirms that the API server is reachable locally, although it doesn't prove that a model can be downloaded.

How to verify: Confirm that the local address responds and run ollama list in your terminal to see whether the CLI can communicate with the server.

2. Check the Model Name and Pull the Correct Reference

If the server responds but downloading a model fails, check the model reference carefully.

Open PowerShell, Command Prompt, or your terminal and run:

ollama list

This shows models already available locally. It doesn't prove that every other model reference is valid.

Next, check the model name and tag you're trying to download. For example, a commonly used model reference is:

ollama pull llama3.2

Use the exact model name and tag documented for the model you intend to install. Don't assume that adding a random version number will work.

If you're following a tutorial, compare its command with the current model reference. Older tutorials may use names or tags that have since changed or become unavailable.

Also, distinguish an invalid reference from a network error. A mistyped model name won't be fixed by restarting your router.

How to verify: Run the corrected ollama pull command and confirm that Ollama begins downloading the model layers rather than immediately reporting a manifest error.

3. Test Port 11434 and Check Whether the Server Is Listening

If the browser still reports a connection failure, check the local port directly.

On Windows, open PowerShell and run:

Test-NetConnection 127.0.0.1 -Port 11434

Look at TcpTestSucceeded. A value of True means a connection to that port succeeded. False suggests that no reachable service is accepting the connection at that address and port.

You can also check whether a process is listening:

Get-NetTCPConnection -LocalPort 11434 -State Listen

If no listener appears, restart Ollama and test again.

You may also try starting the server manually:

ollama serve

Important: The desktop application may already have started the server. If the command reports that the address is already in use, don't keep launching additional instances; check the existing process instead.

If you changed Ollama's host configuration, check whether the server is listening on the expected address. Avoid exposing the API to your local network unnecessarily.

How to verify: Run the port test again and confirm that TcpTestSucceeded is True before retrying the application or model pull.

4. Check Internet Access, DNS, Proxy, and Firewall Settings

If the local API works but the model manifest still can't be downloaded, investigate the connection between your computer and the model registry.

First, open a normal website in your browser and confirm that your internet connection works. If it doesn't, fix that problem before troubleshooting Ollama.

Next, check whether a VPN, proxy, or managed network is interfering with downloads. If you're allowed to do so, temporarily disconnect your VPN and retry. Don't disable corporate security controls or your firewall permanently.

On Windows, open:

Settings > Network & internet > Proxy

Check whether a manually configured proxy is expected. In a terminal, you can also inspect proxy-related environment variables:

Get-ChildItem Env: | Where-Object {
    $_.Name -match 'PROXY|OLLAMA_HOST'
}

If a proxy is incorrectly configured, correct it according to your network setup rather than deleting every environment variable.

If the computer is using custom DNS or a restricted network, try a trusted alternative connection, such as a phone hotspot, if your data allowance permits it. A successful pull on another network points toward the original connection or its restrictions.

How to verify: Retry the same valid ollama pull command and check whether the manifest can now be retrieved.

5. Check Disk Space, Logs, and Ollama Configuration

If the first four fixes haven't worked, inspect the installation before deleting anything.

Check the free space on the drive where Ollama stores its models. Large models can require substantial storage, and the final space needed depends on the specific model and its format.

On Windows, open File Explorer and check This PC. If the drive is nearly full, free space carefully before attempting another download.

Next, check Ollama's logs for details about connection failures, registry requests, or storage errors. On Windows, logs are commonly available under:

%LOCALAPPDATA%\Ollama

You can press Win + R, paste that path, and press Enter. Available log files vary by version.

Also check whether OLLAMA_HOST has been set to an unexpected address or port. Incorrect configuration can make the CLI or another application look for Ollama in the wrong place.

Avoid deleting the entire Ollama model directory or removing downloaded layers just to clear the error. That can force large downloads again and may destroy local model data. If the logs show a specific failing request, use that information to narrow down the problem.

How to verify: Confirm that the model directory has enough free space, review the latest relevant log entries, correct any identified configuration issue, and repeat the pull.

What to Do If Nothing Works (Software vs. Hardware Failure)

Ollama errors are usually related to the server process, network, configuration, or model download rather than failed hardware.

Likely server or configuration issue: The API doesn't respond on port 11434, but restarting Ollama restores it. Check the background process and host configuration.

Likely registry or network issue: The local API works, but valid model pulls fail. Compare results on another permitted network and inspect the relevant error messages.

Likely storage issue: Downloads repeatedly fail when the drive is nearly full, or Windows reports disk errors. Check drive health and make sure the model directory is accessible.

Possible application integration issue: ollama list and model downloads work in the terminal, but a separate local AI interface reports connection refused. Check that application's API URL, usually set to http://localhost:11434, and confirm that it uses the correct port.

Reinstall Ollama only after checking these points. Before uninstalling, preserve any local models you need and make sure you understand where Ollama stores them on your system.

Common Questions From Users

Why does Ollama say "connection refused" on localhost:11434?

The local Ollama server usually isn't listening at that address and port, or the application is using the wrong endpoint. Restart Ollama, test the local API, and check port 11434.

How do I fix Ollama "Error: pull model manifest"?

Verify the model name and tag, confirm that the local server responds, then check internet, proxy, and registry access. If the reference is invalid, changing local server settings won't fix it.

Why does Ollama work in the terminal but not in my web UI?

The web interface may be pointing to the wrong address, port, or host. Confirm that Ollama is running and that the interface is configured to use the correct local API endpoint.

Should I reinstall Ollama to fix a manifest error?

Not as the first step. A reinstall won't fix a misspelled model reference, a blocked registry connection, or an incorrect API URL. Diagnose the failing layer before removing the installation.

Bottom line: Check the local server first, verify the model reference second, and investigate network access only when the server itself is responding. That sequence helps you isolate the problem without losing downloaded models or reinstalling software unnecessarily.