Fix Nanobrowser errors and problems

Find the message you see, then follow the steps under it. The messages are grouped by where they show up:

Setting up a model
Errors from your model provider
During a task
Tabs and sites
Files
Voice input
Pages the agent can’t use, what isn’t supported yet and upgrading from v0.2

`No model is chosen. Add one in Settings → Providers.`

Nanobrowser has no model to work with. The side panel may show the same thing as a card: Choose a model to start.

Open Settings → Providers.
Add a provider, and a model for it.
Choose that model in the side panel.

`Connected, but the model did not call the test tool…`

The model answered, but it didn’t call the test tool, so it’s marked No tool calling.

On the welcome page, the same problem reads … answered but did not call the test tool, so it cannot run the agent. Choose another model.

Choose a model that supports tool calling. See Test models.

`The reply was cut short before the model called the test tool…`

The model used up its reply, often on reasoning, before it called the tool. Run Test models again.

`Ollama refused the request from the extension…`

Most often, Ollama doesn’t accept requests from browser extensions yet.

Set OLLAMA_ORIGINS to chrome-extension://*, then quit and reopen Ollama. The message shows the steps for your system. See Use Ollama.
Check that Ollama is running.
Check the Server address. It’s usually http://localhost:11434.

`LM Studio refused the request…`

Turn on Enable CORS in LM Studio’s Developer settings, then try again. See Use LM Studio.
Check that the LM Studio server is started.
Check the Server address. It’s usually http://localhost:1234.

`The server refused the request or could not be reached…`

You see this with OpenAI-compatible, xAI Grok, DeepSeek and Mistral.

Check that the server is running.
Check the Server address. Nanobrowser uses it exactly as you type it, so include /v1 if the server needs it.
Allow CORS requests from browser extensions in the server’s settings.

See Other servers and OpenAI-compatible.

`Could not reach …`

Nanobrowser couldn’t connect to the address in the message. Check your connection, and any VPN, proxy or firewall.

`The server lists no models…`

Your local server is running, but it has no models yet. Download one first, for example with ollama pull. Then click Connect again on the welcome page, or List models in Settings → Providers.

`This model has a 4,096-token context window…`

Tasks on real pages need 16k tokens or more. Set a larger context in Ollama or LM Studio, then load the model again. See Run a local model.

Your API key is not accepted

You see The provider did not accept this key. Check it and save it again., or the provider’s own message, often with 401 or invalid API key.

Copy the key again from the provider’s site, and check that it belongs to this provider.
In Settings → Providers, point at the saved key and click the cross (Clear key).
Paste the new key, then press Enter or click the check mark.

Some providers also refuse a valid key when your account has no credit left.

Rate limit or quota errors

The message often has 429, rate limit or quota in it. Nanobrowser retries a few times before the task stops.

Wait a minute, then click Retry.
Check your plan, limits and credit on the provider’s site.
Choose another model.

Model not found

The message often has 404, not found or does not exist in it.

Check that the model name is spelled exactly as the provider writes it.
Click List models in Settings → Providers to see the names the provider offers.
The provider may have retired the model. Choose a newer one.

`This conversation no longer fits the model's context…`

The chat got too long for the model, even after compaction. Start a new chat, or choose a model with a larger context window.

`The model's reply was cut before it said anything…`

The model used its whole reply on reasoning. Try again, or set Reasoning lower in the side panel.

`This model did not take the reasoning setting; continuing without it.`

The model doesn’t accept the Reasoning setting. The task goes on without it.

`This model did not take images; continuing without them.`

For tasks that need screenshots or attached images, choose a vision model.

`The agent stopped at its turn limit.`

Click Continue to let it go on. To change the limit, see Maximum turns.

`The run was interrupted.`

The browser or the extension restarted during the task. A step that was running may or may not have happened, so check the page, then click Resume.

`The run failed: …`

An unexpected error. The text after the colon says what. Try again with Retry. If it keeps happening, open an issue on GitHub.

`That tab is on a site you blocked, so it was not shared.`

The tab’s site is set to Block. To change that, see Site rules.

`The other conversation is working in that tab. Stop it first.`

Another chat’s task is using that tab. Stop that task, then move the tab.

`The shortcut's start page is on a site you blocked. Nothing was sent.`

The shortcut opens a site set to Block. Change the shortcut’s Start page (optional) in Settings → Shortcuts, or change the site’s rule in Settings → Permissions.

`…: PDF and Office files cannot be read in this version…`

The agent can’t read PDF, Word, Excel or PowerPoint files yet. Copy the text into a text file and attach that, or paste the text into your message.

A file can’t be attached

…: only images (PNG, JPEG, WebP, GIF) and text files can be attached.
… is larger than 25 MB.
… was not attached: a conversation holds at most 100 MB of files.

See Attach files for what you can attach. For more files, start a new chat.

`This model cannot see images: … is kept with the conversation but not sent.`

Choose a vision model, then send your message again.

`Saving needs your permission in Chrome. Press Save to allow it, or Skip.`

Click Save, then allow it when Chrome asks. You can also allow it in Settings → Permissions → Additional browser permissions.

`To talk instead of typing, choose a voice model in Settings.`

Add an OpenAI or Google Gemini provider in Settings → Providers, then choose a Voice model there. See Use voice input.

Microphone problems

Nanobrowser can't use the microphone. Allow it in Settings → Permissions.: open Settings → Permissions and allow the Microphone. If you blocked it before, allow it again in the site settings, from the icon left of the address bar.
No microphone was found. Connect one and try again.: connect one, and check that your system lets the browser use it.
No speech was heard. Try again, a little closer to the microphone.: check that the right microphone is chosen in your system settings.

`Could not turn speech into text: …`

The text after the colon is the voice provider’s error. Check your key and credit with that provider.

Pages the agent can’t use

The agent can’t read or act on:

Browser pages: chrome:// pages in Chrome and edge:// pages in Edge, such as settings and the extensions page.
Pages of other extensions.
The Chrome Web Store and the Edge Add-ons store.
Files on your computer (file: addresses).
PDFs open in the browser’s PDF viewer.
Sites you set to Block in Settings → Permissions.

If the agent says it can’t use a page, open a normal web page and try again. To give it the text of a PDF, copy the text into your message.

What isn’t supported yet

Browsers other than Chrome and Edge. Other Chromium browsers may work, but they aren’t tested.
PDF and Office files as attachments. Paste the text instead.
Voice input with other providers. Only OpenAI and Google Gemini voice models work.
Azure OpenAI. It isn’t one of the providers in this version.
Letting the agent open only certain sites. You can block sites, but you can’t limit the agent to a list.

Upgrading from v0.2

When Nanobrowser updates from v0.2, it moves what it can:

Your providers and API keys. Azure OpenAI isn’t moved, because this version doesn’t support it.
Your old Navigator model and its reasoning effort, which become your default model.
Your saved prompts, which become shortcuts. Type / in the side panel to use them. A /summarize shortcut is added too.
Your blocked sites, if your old firewall was on, as Block rules. A rule no longer covers subdomains, so add the ones you care about.
Your choice to turn off Help improve Nanobrowser. See Privacy and security.

Other settings aren’t moved. An old list of the only sites the agent could open is gone.

After the update, a What’s new in Nanobrowser 0.3 card in the side panel shows the main changes.

Legacy conversations

This version can’t open your v0.2 chats. They stay in your browser until you export or delete them:

Open Settings → Data.
Find Legacy conversations (v0.2).
Click Export as Markdown to save them all in one file.
To remove them, click Delete, then Delete them now.