Open source · Your choice of model
A helpful claw
for your inbox.
Write, rewrite, summarize, and translate in Thunderbird with your chosen AI provider or local model. Review every suggestion and keep the final say.
- Generate
- Preview
- Apply
- Undo
Bring your own model
Your provider.
Your preferences.
Connect directly to a hosted provider or a compatible server on your computer. Choose your model in the extension and make ThunderClaw your own.
- OpenAI
- OpenRouter
- DeepSeek
- Fireworks AI
- Ollama
- LM Studio
- Custom endpoint
OpenAI-compatible connections. Model support for structured output and thinking controls varies. Check compatibility ↗
A little help, right where you write
Find the words.
Keep your voice.
Polish a draft, find the point of a long message, or read across languages. Choose an action when you need it.
Illustrative example · Improve
A clearer draft.
The same intention.
Improve, proofread, shorten, adjust tone, translate, summarize, or give a custom writing instruction. Preview the result before applying it to your selection.
Your selection
could you send the updated schedule by friday? we need it before we can confirm the booking.
Suggested edit
Could you send the updated schedule by Friday? We need it before we can confirm the booking.
You review. You apply. You can undo.
Illustrative example · Summarize
The details that matter.
A separate summary card helps you get your bearings while keeping the original message in view.
Garden workday
- Meet Saturday at 9 AM by the greenhouse.
- Bring gloves and a water bottle.
- If it rains, projects move indoors.
Illustrative example · Translate
Read across languages.
Choose a target language and let Auto-detect handle the source. Translate visible text in place, then restore the original when you want.
English → French
The meeting starts at ten.
La réunion commence à dix heures.
Set up ThunderClaw
One extension.
A connection you choose.
Configure your provider and model directly in Thunderbird. Your connection and API key stay saved on this device across restarts.
Available on Thunderbird Add-ons. See the setup guide for connection settings and local development builds.
-
01
Add the extension
Use Thunderbird 128 or newer. Install ThunderClaw from Thunderbird Add-ons, approve the requested permissions, then open its settings.
Get the extension ↗
-
02
Choose your connection
Select a provider preset or enter a compatible API base. For Ollama or LM Studio, start your local server and load a suitable model.
-
03
Pick a model and save
Enter an API key if required. Click Load models where available and choose one, or enter its exact ID. Fireworks AI requires a full model ID entered manually. Save the connection and approve host access when Thunderbird asks.
-
04
Test, then try it
Click Test connection to check your model with synthetic text. Then select text in a draft, generate a preview, and decide whether to apply it.
Use HTTPS for remote servers. Local HTTP supports explicit 127.0.0.1 or [::1] addresses.
Your inbox, your decisions
Help with the words.
You handle the sending.
Review every edit
Generate, Preview, Apply, and Undo are separate choices. Thunderbird handles sending. Models have no authority to change recipients, attachments, or headers.
Keep supported formatting
Eligible selections on qualified Thunderbird 153+ versions support paragraphs, flat lists, bold, italic, and underline. Received translations keep the message’s existing structure.
Know where your text goes
Actions send selected content and necessary context directly to your chosen endpoint. Provider privacy policies apply. Local servers and routing services can use upstream cloud models.
Read the privacy policy ↗
A few useful details
Before you get started.
Will I have to enter my API key every day?
Your key is saved in your local Thunderbird profile and survives restarts. It is never synchronized. Local profile storage is not an encrypted vault, so protect your device and backups. Remove the saved key in settings; revoke it with your provider if needed.
Can I use a model running on my computer?
Yes. Ollama and LM Studio have presets, and you can configure another OpenAI-compatible server. The model must support the selected structured output. Check your server’s model settings if you need inference to stay on your computer.
Can I control response length and thinking?
Yes. The response-token limit defaults to 16,384 and helps bound generation cost and waiting time. Thinking defaults to Off for DeepSeek and Model default for other presets. Available controls depend on the provider and model. Save and retest after changing them.
What does ThunderClaw cost?
ThunderClaw is open source under the Apache 2.0 license. You supply the model connection. Hosted providers may charge for generation, including connection tests; a model running on your computer uses your own hardware.