Typli update: web search, new models, and more

We've been heads-down shipping for the last few months and haven't really talked about any of it. Time to fix that.

Here's what's new in Typli.

Every generator can search the web now

This was the one that bugged me the most. You'd ask a generator to write about something happening right now and it would confidently make things up, because it only had its training data to work with.

That's fixed. Every writing tool on Typli now searches the web automatically when it needs current information. You don't toggle anything on. The AI figures out when it needs to look something up, runs a search, and pulls in what it finds.

If you ask the blog post generator to write about something that happened last week, it'll search for it first. Same with the article rewriter, the email generator, and all the others.

The sources show up in a collapsible section under the output, so you can check where the information came from and link back to it if you want.

More models

When I started Typli, we had one model. Now there are ten, and you can switch between them on any generator.

Here's the current lineup:

GPT-5.2 Chat is the default. It's fast and handles most writing tasks well.
GPT-5.4 is OpenAI's newer, more capable model. It writes better long-form content and handles complex instructions more reliably. Subscriber only.
Gemini 2.5 Flash has a 1M token context window, which is wild. Good for working with long documents.
Claude Haiku 4.5 from Anthropic. I like its writing style for certain things, especially when you want something that reads less like a template. Subscriber only.
Grok 4 Fast and Grok 4 Fast Reasoning from xAI. The reasoning variant shows you its thought process, which is useful for complex prompts. Both subscriber only.
DeepSeek Chat and DeepSeek Reasoner are affordable options. The reasoner shows its thinking, similar to Grok's reasoning mode. Reasoner is subscriber only.
Kimi K2.5 is strong on reasoning and coding. 262K context window.
Qwen 3 235B from Alibaba. Surprisingly good for the price.

All of them have web search. That was the whole point of building it as a separate tool rather than relying on each provider's implementation.

You can watch the AI think

Some of these models (DeepSeek Reasoner, Grok 4 Fast Reasoning) do chain-of-thought reasoning before they answer. We used to hide that entirely. Now you can see it.

While the model is working through your prompt, a "Thinking..." indicator shows up. Once it starts writing the actual answer, the thinking section collapses. You can expand it later if you're curious about how it got there.

I find it genuinely useful for debugging bad outputs. If the AI misunderstood your prompt, you can usually see where it went wrong in the reasoning.

New generators

We added a couple of email tools:

Follow-up email generator for general follow-ups
Follow-up after interview generator for post-interview thank-you emails

These use the same engine as everything else, so you get model selection and web search on both.

The interface got cleaned up

This is harder to describe in a blog post, but the generator pages look better. We tightened up the spacing, put the model selector and generate button on the same row, and made the output area feel less cluttered.

Nothing about how the tools work changed. It just looks more polished.

iOS app is coming

We've been working on a native iOS app. It's built and functional, but we're still going through the App Store review process. Stay tuned on that one.

What's next

We're adding models as they come out, finishing up the iOS app, and continuing to improve the writing tools. If something is broken or you want a feature, use the feedback button on any generator page.

Try the updated generators here.