Launched this week

Teaflask
Self-improving AI copilot for your app. No backend changes.
259 followers
Self-improving AI copilot for your app. No backend changes.
259 followers
Turn your existing SaaS into an AI-native product without rebuilding your backend. Teaflask adds a managed AI copilot that lets users get things done by simply asking. Integrate through your frontend, use your own UI components, and go live in a day. Teaflask keeps its product knowledge grounded in your code, learns from real usage, and lets you test changes before release.










Free Options
Launch Team

AWS Startup AdvisorPersonalized AWS guidance that meets you where you’re already building
Promoted
Hello hello, Ragav from Teaflask here!
We’re building Teaflask to make the SaaS product you already ship agentic. Your users describe what they want done, and an embedded assistant does the work inside your product.
What it is
A self-improving and ready-to-deploy assistant for your frontend (or slack), backed by a managed agent runtime.
It can answer product questions, navigate, fill forms and call your APIs. With tool views, your own components render results inside the conversation, so it feels like your product, not a chatbot bolted onto it.
How it works
1. Connect your repository (optional but recommended) to help prepare product documentation and discover API actions.
2. Configure your agent and set each action to Allow, Ask or Deny.
3. Embed the assistant and connect it to your existing API client (there is a prompt on our landing page which you can give to your coding agent).
We host the agent runtime. Product API calls happen in the user’s browser through their existing session, keeping your backend’s authentication, permissions and business logic in place. No backend changes required.
How it keeps improving
Teaflask checks documentation claims against your code and flags drift. Steep examines user journeys and assistant conversations to suggest improvements, backed by evidence you can hand to your coding agent.
Before shipping changes, replay saved conversations to check for regressions or see whether a cheaper model can do the same job. Your team controls what ships.
Getting started
We have Free and Enterprise plans. Most teams should be fine on Free; please book a call if you want to scope it for enterprise. For LLM usage, you can bring your own model keys, or fund your Teaflask wallet (at 0% markup).
Would love to see you in our Discord!
Congrats on the launch, Ragav!
This seems really interesting. How does the configurability look? After all, it's inevitable that I can control more things if I build using my own API key, right?
@peter__ishere Thanks!
Most of what you’d want to tweak is configurable. You pick the model (Anthropic, OpenAI or anything on OpenRouter, on your own key), write the prompt, and write down skills for it to follow. Every action gets Allow, Ask or Deny. You can plug in your own MCP servers for custom tools, add guardrail rules, give it memory etc.
What you get on top is the part I’d hate to build again. Runs are durable, so a crash, a deploy or an approval that sits for two days doesn’t lose anything. Streaming picks up mid-sentence after a dropped connection. Real conversations get saved as test cases, so when you change the prompt or the model you can see what broke before it ships. Docs get tested against your actual code, so when the code changes you hear about it. And there’s a regular report on where users are getting stuck.
What you can’t change is the agent loop itself, meaning the code that calls the model, runs the tools and manages context. That’s ours. If you need control at that level, you’re honestly better off building on the API yourself:)
@ragavsachdeva Thanks so much. will explore/compare some other options, but Teaflask is definitely on the top of my mind.
This seems so good for legacy products!! What level of customization do you allow people on the UI components?
Thanks@agarwalml! You can customize as much or as little as you want:
Out of the box: four layouts for the same conversation: a corner widget, a sidebar, a ⌘K command palette and a full page. Light and dark modes are included.
Theming: set the colors, corner radius and font, either as a prop or with CSS variables. You can also swap in your own logo or mascot and your own opening prompts.
Your own components in the chat: when the agent runs an action, you can show the result with your own component, like an order card instead of raw JSON.
Fully headless: use hooks for the transcript, the message box and approvals, and build the entire UI yourself. We still run the agent underneath.
Congrats on the launch!
Our ID tokens last five minutes and we sign people out when theyre idle. an approval sitting for two days outlives the session thats meant to run it.
If that session is gone when the user finally approves, what makes the call?
tool views rendering your own components inside the chat is clever Ragav, congrats!
Thanks @mohitvishwakarma
This is really an interesting idea. I should try this on my own product.
How can I add this one to my existing product?
@helloshams There is a setup prompt on teaflask.com that you can give to your coding agent and it should get you started:)
Love the no-backend-changes angle — that's the part that would actually get me to try it on a shipped product. A copilot grounded in my code, that can answer questions, navigate, fill forms and call my existing APIs without me rebuilding anything, is a much easier sell than another chatbot bolted on. The self-improving bit — replaying real conversations to catch regressions — is the detail that makes me trust it more.