Connect with the Product Hunt community

Thanks a lot for supporting us and for your feedback with SKI.

We have taken in your feedback and requests. We have transformed SKI into a voice-first coding harness, with enhancements to support multi modal inputs in realtime

• Supports 85+ languages: You could speak to your coding agents in languages that you are comfortable in, and the agent will talk to you in the same language as well.

• Prompt processing: It runs your audio and brings in your thoughts as a proper prompt, refined for most agents (with optional approval)

• All agents in one UI: Brings in all agents into a single UI, with most commonly used tools for coding. Switch sessions and agents by voice.

• Chained agents: With a single voice command, you can give the agent a chained task. For example, ask Opus 5.5 to plan, and then to hand it over to GPT Luna to code and then to Fable for review, in just a simple voice command

• Intelligent router: With "One chat for all" turned on, you can just talk to a single agent and it will intelligently route to sessions (with optional approval)

• Expressive voice: A more human-like, low latency voice for each sessions with 28+ voices

One of the tricky parts of building voice agents is handling conversations that don t get resolved in a single call.

A customer might be busy, ask for a callback tomorrow, or simply not answer the first time. We wanted to make sure these scenarios are handled as part of the voice workflow, not left for someone to manage manually.

At Vaani AI, we ve built callbacks and retries into the platform.

Here s how it works:

A voice agent isn t just about what it says. It s also about how it says it, how quickly it responds, and whether it can adapt to the flow of a conversation.

We ve been working on making voice behaviour more configurable at Vaani AI. Here s what you can tune:

• Speaking speed: Adjust the pace of the agent s voice depending on the TTS provider.

• Response timing: Control how quickly the agent responds after the caller stops speaking, and experiment with speculative response generation for faster turn-taking.

• Emotional expression: Configure supported emotions and voice tags so the agent can sound excited, calm, or more expressive where appropriate.

• Voice consistency: Tune parameters like stability and style to balance a consistent voice with more natural expression.

• Volume control: Callers can ask the agent to speak louder, quieter, mute, or restore the volume during a call.

• Natural audio details: Add optional breathing sounds and background ambience to make conversations feel less robotic.

• Entity handling: Give the agent more patience when callers are spelling out names, phone numbers, or other details.

One thing worth mentioning: these controls depend on the TTS provider. The same configuration won t necessarily produce the same behaviour across different voices.

A lot of teams already manage their leads and customer data in Google Sheets. Exporting that data, uploading a CSV, and then figuring out how to get call results back into the same workflow adds unnecessary steps.

So we built a direct Google Sheets integration into Vaani AI Campaigns. @Vaani Developer Platform

Here s what you can do:

• Connect your Google account: Pick a spreadsheet and the specific tab you want to use.

• Map your columns: Map phone numbers, names, and agent variables like city or plan type to Vaani s contact fields.

• Control what gets written back: Choose columns for call summaries, status, disposition, duration, and extracted information.

• Launch your campaign: Vaani picks up contacts from the sheet and starts calling.

• Keep it in sync: Once calls are processed, their results are written back to the corresponding rows. Add new rows later and they can be picked up on the next sync without importing everything again.

One thing that s easy to overlook when building voice agents is what happens when a provider goes down or a model starts returning errors mid conversation.

In a normal API workflow, a failed request might mean retrying or showing an error. In a live voice call, it s different. The caller is waiting, and a failure in one component shouldn t bring down the entire conversation.

At Vaani AI @Vaani Developer Platform ,we ve been working on a multi-layer failover mechanism to handle this:

• Independent fallback chains: Speech recognition, LLM and TTS each have their own backup providers.

• Error-based switching: If the primary provider returns an error, the request moves to the next configured provider.

• Safety-net fallback: Even when an agent is misconfigured, there s a fallback model available.

• Configurable routing: Teams can choose primary and backup models, use their own API keys, and control whether fallback is enabled.

You do not write the prompt, you teach it: build and fix voice agents by chatting

Writing a good agent prompt is a skill most teams do not have.

The last two days, I had some personal problems and spent basically the whole time trying to solve them.

And during that time I did no work, no posts, no ideas, no inspiration, nothing... =D

Just me, solving my own puzzle !

Hi everyone!

Models give an agent a voice. Behaviour is what makes it good on a real call. That s the part Vaani Developer focuses on.

In the platform you set up:

• The call flow, from opening to wrap-up.

• How it handles objections, interruptions and edge cases.

• Actions and tools: hold the call, schedule a callback, transfer to another agent or to a human.

• Conversation settings, background noise and provider choices.

Not just use AI for moodboards or a few shots but make it part of the normal production workflow.

What has to change first: quality, control, client trust, rights, pricing, or something else?

I have heard/read that people started making mistakes in their text to sound more like a human and less like an AI.

+ There was a discussion that LinkedIn identified AI-generated text + Claude released text watermarks.

I run a daily checklist for my social accounts, and this week I pulled the numbers to find out why Reddit had gone flat. Comment karma sits at 125. My last post there was 20 September, 47 days ago.

The strange part is that I already knew posts were the lever. My best post drew 62 comments. My best comment ever scored 8. On 20 September I wrote that into a plan: post 2 to 3 times a week. On 17 October a second review said it again: one post this week.

I recently got really excited about an idea and started working on it, only to realise that it's already been done. Quite a few times, actually.

Sure, there are always ways to differentiate, find a niche, or approach the problem differently. But it got me thinking: At what point do you decide a market is too saturated to enter?

Your AI can see your screen. Now imagine it drawing directly on it.
Today, we re introducing on-screen drawings in Aura.

As developers and creators, we spend half our day switching between browser tabs, heavy task managers, and notifications. I wanted something fast, beautiful, and native-feeling that sits right on your desktop keeping your core tasks visible without getting in your way.

That s why I created Dots.

Why you might like it:

• Clutter-free workflow: Keep your primary focus areas visible without context-switching.

• Designed for speed: Quick input, clean layout, and zero unnecessary friction.

• Polished aesthetic: Subtle visual touches that complement your workspace.

Lancetta 0.64.0 is out: Lancetta is on your desktop now.

A small widget for Claude and one for Codex, with both windows and the time to each reset, and a medium one with every agent and the suggestion that matters most.

I build by directing AI agents. I'm not a developer, so tests are how I know something works. The app has about 3,600 of them. They run three times before anything gets merged. One day a deploy went out before its database migration. The new code went live expecting database changes that hadn't run yet. The app was down for 5.5 hours. Every test had passed. That's the part I keep thinking about. The tests run against development, not production. They check that the code does what it should. Nothing checked the order things shipped in. So a green suite told me the code was fine. It said nothing about the deploy. If you ship with AI agents, what checks the deploy itself? Migrations, order of steps, anything that only exists in production?

We launch a feature here on Thursday, and the landing page section for it is the one new users keep bringing up on calls and in replies, more than the hero, more than pricing.

It is a simple block. One sentence: see the change before it ships. One picture of a working copy of your site with the change on it, and two buttons: approve, or send it back.

I think it works because it answers the question people actually have about any tool that writes code for them: what if it gets it wrong? The answer is not trust us , it is look at it first . The other sections explain what it does. This one removes the fear.

Curious what the equivalent section is on your landing page: the one that answers the fear rather than describing the feature. And for those who have run the test, did it convert better than the feature description?

Hi everyone!

We keep getting asked what we mean by orchestration platform . Here s the simplest way we can put it.

Think about paying on a big shopping site during a sale. If one payment gateway fails, the system quietly retries on another one, and you never see the failure. Sites do this because every failed payment is lost revenue.

Voice AI works the same way:

Nothing frustrates a customer more than repeating themselves on every call.

Memories lets your agent recall what happened in earlier conversations with the same person, so a returning caller does not start from zero.
In @Vaani Developer Platform you can choose how many past calls it remembers. Let one agent see the history of related agents, so the handoff between them feels continuous. also bring in your own customer details from your systems, such as plan, last order or account status from APIs

The result is calls that feel like a conversation with someone who knows you.

What should an agent remember about a returning customer?

@akash_anil254 asked a few days ago what my strategy is for Product Hunt (I do not want to say I do not have one, but I really do not have one). I enjoy being present here.

Being here already has its own benefits:

• Have found a supportive community with an entrepreneurial mindset

• I am well-informed about the latest tech trends

• Building my personal brand

A call ends and then what?
Someone has to copy the outcome into a spreadsheet, a CRM or a chat channel.

Custom webhooks send an update to the address you choose when something happens: a call starts, ends, is not answered, fails or is transferred.

• Choose the events you care about.

• Works with no-code tools like Zapier, Make and n8n.

• Updates carry the phone number, so you can match them to your customer records.
  All of this only on @Vaani Developer Platform

What should happen automatically after a call ends?

Lingo.dev can now automatically provision your localization engine s context from your website: generate a glossary, translation rules, and a brand voice, then save and attach them to the engine.

When creating an engine in Infrastructure, enter your website. Lingo.dev reads the content and creates the context automatically, ready for you to review and refine. Everything stays editable. Try a representative translation in Playground before using the engine in your integrations.

For an existing engine, select Fill missing context, enter your website, and choose which missing types to generate. Types already attached to the engine are left alone. Existing context isn t replaced.

A voice agent that makes up answers is worse than no agent at all. Callers ask about prices, policies and product details, and a wrong answer costs trust.

With Knowledge Base, you can upload your files, paste text, or point it to your website. During a call, the agent looks up the answer in your material and uses it.

• Works with long, multi-page documents.

• Add or update content without rewriting the agent.

• You can switch it off for agents that do not need lookups.

A short pause can happen while it looks something up. The answer is worth the wait.

I was 15 when I entered university...

In my 4th year of engineering, I dropped out to go all-in on Murmell.

We re giving AI tools and other tech companies more and more data.
And it s not just our names and email addresses.

Sometimes, it gets a little too personal:

Phone numbers. I hate getting SMS messages, especially since I only use my phone when absolutely necessary. (Family calls and important things related to doctors)

Biometric data. Like fingerprints.