Kollect Voice Agent: An Open-Source Experiment Turning Forms into Voice Conversations

2026-02-20 | ProductHunt | Official Site

30-Second Verdict

What is it?: It turns traditional questionnaires into voice conversations. Instead of typing into boxes, you talk directly into the microphone, and the AI guides you through the survey in real-time. You can even create forms just by describing them in natural language.

Is it worth your attention?: If you're an indie developer, it's worth 20 minutes to check out its tech stack and implementation. As a standalone product, it's very early (only 4 PH votes, launched 2 days ago), but the "voice-replacing-forms" direction it represents is definitely taking off in 2026.

Three Key Questions

Is it for me?

• Target Audience: Product teams, market researchers, and UX researchers who need to collect user feedback.
• Are you one?: If you're tired of sending surveys with soul-crushingly low response rates, or if your users hate typing (like the elderly or blue-collar workers), you are the target.
• Use Cases:
  • Customer Satisfaction: Replace cold forms with voice chats; completion rates can jump by 45%.
  • Scalable User Interviews: Let AI conduct semi-structured interviews for you, saving massive manual labor.
  • Accessibility: For visually impaired users or those who find typing difficult, speaking is much more convenient.
  • Internal Quick Polls: You probably don't need this; Google Forms is enough.

Is it useful?

ROI Judgment: Currently not recommended for formal production projects. However, if you're a developer wanting to study how to build "Voice + AI + Forms," this is excellent material—modern tech stack, clear architecture, and fully open-source.

Is it a crowd-pleaser?

The Sweet Spots:

• Create forms with your voice: No dragging fields around. Just say, "I want to make a 5-question customer satisfaction survey," and the AI generates it.
• Voice filling: Finally, no more squinting at tiny input boxes on a phone; just speak.

The "Wow" Moment:

"Filling forms can be extremely boring, so why not just talk to them?" — The developer's own words in the Vercel community.

Real User Feedback: The product just launched 2 days ago (approx. 2026-02-18) with only 4 votes on PH and zero discussion on Twitter/Reddit. The developer posted "Would love your feedback" in the Vercel community and is waiting for the first batch of seed users. To put it bluntly, if you try it now, you're an alpha tester.

For Indie Developers

Tech Stack

This is a very modern full-stack combo:

• Frontend: Next.js (React full-stack framework)
• Backend/Database: Convex (Real-time reactive database with serverless functions)
• Speech Recognition (STT): Deepgram AI (Real-time voice-to-text, latency <300ms)
• LLM Routing: OpenRouter (Unified interface to access 500+ models)
• Auth: Clerk (Free up to 50,000 MAU)
• Deployment: Vercel

The interesting choice here is using OpenRouter instead of calling OpenAI directly. This allows for flexible switching between underlying models (GPT-4o, Claude, Gemini) without vendor lock-in. Using Convex for a real-time database is also smart—voice conversations require real-time sync, and Convex natively supports sub-50ms latency.

Core Implementation

Technical breakdown (based on the stack):

1. User speaks -> Deepgram performs real-time STT.
2. Transcribed text goes to an LLM via OpenRouter -> Understands intent and extracts answers.
3. LLM dynamically generates the next question based on form structure -> TTS plays it (or text displays).
4. Conversation data is synced to the Convex database in real-time.
5. Form Creation: User describes needs -> LLM generates the form schema.

The challenge here is latency control. The entire loop (input -> recognition -> LLM thought -> response) needs to be within 1-2 seconds to feel natural. Deepgram's API is fast (<300ms), but LLM inference remains the bottleneck.

Open Source Status

• Is it open source?: Yes. PH tags it as Open Source, and Vercel posts mention it's "fully open source."
• GitHub Repo: I searched but couldn't find a public link yet. The developer's GitHub is admildo; he previously made 'speechy,' but the Kollect repo might be private or under a different name.
• Similar Projects: Formbricks (Open-source forms, no voice), Vocode (Open-source voice AI framework).
• Build Difficulty: Medium. If you know Next.js + Deepgram + LLM APIs, one person could build an MVP in 2-3 weeks. The difficulty lies in polishing the UX, not the tech.

Business Model

Currently none; it's an open-source side project. API cost reference:

For light use (a few hundred surveys/month), the free tiers will likely cover everything, making the total cost $0.

Giant Risk

High. Google Forms has AI features, Typeform is adding AI, and Microsoft Forms will follow. However, these giants focus on "AI-assisted form generation," not "voice replacing forms." Real competition comes from vertical players like TheySaid and Voiceform, who already have funding and paying customers.

For Product Managers

Pain Point Analysis

• Problem Solved: The two chronic diseases of traditional forms—low completion rates and poor response quality.
• How painful is it?: Very. Data shows up to 80% of online surveys are abandoned. Response rates dropped from 40% in 1997 to less than 5% today (Pew data). Voice surveys can pull completion rates from 28% to 76% (in education) and capture 67% more emotional data.
• Frequency: Medium. You don't do surveys every day, but when you do, the low response rate is a major headache.

User Personas

• Market Researchers: Need to collect feedback at scale but facing declining response rates.
• UX Researchers: Want to do interviews but lack the manpower; need AI to scale.
• Product Teams: Tired of receiving perfunctory "it's okay" responses on NPS surveys.
• Educational Institutions: Assessing students where voice is more natural than text.

Feature Breakdown

Competitor Comparison

Key Takeaways

1. "Create with your voice": This is a great interaction idea. Describing needs -> AI generating structure has a much lower barrier than drag-and-drop editors.
2. Voice-First Experience: While others add AI to forms, Kollect tries to replace the form with voice entirely—a more aggressive and interesting path.
3. Tech Stack Choice: OpenRouter for LLM routing + Convex for real-time data is a combo worth noting for future projects.

For Tech Bloggers

Founder Story

• Founder: Admildo Manuel
• Background: Describes himself as "a nerd/Christian with a strong drive for engineering, computer science, data science, math, and entrepreneurship." He writes technical articles on Medium and created 'speechy' on GitHub, showing a long-term interest in voice tech.
• Socials: GitHub / Medium / Instagram @admildofm
• Motivation: You can see the evolution from "Text-to-Speech" (speechy) to "Voice-replacing-forms" (Kollect). The goal seems to be bringing voice interaction into more daily scenarios.

Discussion Angles

• Angle 1: "Is the Form Dead?": Voice AI survey completion rates are 45-170% higher than traditional forms. That's a headline. But the counter-argument is: voice isn't for every scenario (public places, private data, structured data collection).
• Angle 2: Indie Dev vs. VC-backed Rivals: Kollect is a one-man open-source project vs. funded competitors like TheySaid. Is this niche "small and beautiful" or "winner-takes-all"?
• Angle 3: Redefining Forms in the AI Era: Is using 1990s-style HTML forms in 2026 just outdated?

Hype Metrics

• PH Ranking: 4 votes, almost no heat.
• Twitter Discussion: Zero. No one is talking about it yet.
• Vercel Community: Posted 2 days ago in Showcase and Open Source sections.
• Search Trends: Too new to have a trend.

Content Advice

• What to write: Don't write about Kollect alone (too little material). Write about the trend of "Voice AI killing traditional forms" and use Kollect as a case study.
• Trend-jacking: The voice AI sector is hot in 2026 (85% of enterprises plan to adopt it). Combine this with Deepgram and Voiceform for a sector overview.

For Early Adopters

Pricing Analysis

Quick Start Guide

• Setup Time: ~30-60 minutes for developers (env + API keys). Non-developers cannot use it yet as there is no hosted version.
• Learning Curve: Medium (requires knowledge of Next.js deployment and API configuration).
• Steps (Estimated):
  1. Clone the repo.
  2. Get API keys for Deepgram, OpenRouter, Convex, and Clerk.
  3. Configure environment variables.
  4. Run npm install && npm run dev.
  5. Open in browser, allow mic, and start.

Pitfalls & Complaints

1. Missing GitHub Repo: Claimed to be open source but the repo is hard to find. This is the biggest red flag.
2. Environmental Noise: In noisy areas, recognition accuracy can drop by over 30%.
3. Extremely Early: No docs, no community, no case studies. If it breaks, you're reading the source code.
4. Privacy Concerns: Voice data is processed via Deepgram's cloud API.

Security & Privacy

• Data Storage: Convex cloud (can be self-hosted).
• Voice Processing: Deepgram cloud API (SOC 2 Type II, HIPAA compliant).
• Auth: Clerk (SOC 2 compliant).
• Privacy Policy: None for the project itself; security depends on your deployment.

Alternatives

For Investors

Market Analysis

• Conversational AI Market: ~$18-21B in 2026, CAGR 20-23.7%.
• Online Form Builder Market: ~$764.5M in 2026, projected $1.77B by 2035.
• AI Voice Market: $4.16B in 2025 -> $20.71B by 2031, CAGR 30.7%.
• The Intersection: "Voice AI + Forms" sits at the cross-section of two growing markets with clear potential.

Competitive Landscape

Timing Analysis

• Why now?: In 2026, Deepgram's latency is <300ms, OpenRouter makes LLM costs negligible, and Convex provides out-of-the-box real-time sync. This stack wasn't possible three years ago; now one person can build it.
• Tech Maturity: Underlying tech is ready; the bottleneck is product experience.
• Market Readiness: 85% of enterprises plan to use voice AI, but the "voice form" niche is still in the education phase.

Team & Funding

• Founder: Admildo Manuel, Indie Developer.
• Core Team: 1 person, Solo project.
• Funding: None.

Investment Perspective: Kollect itself isn't an investment target, but the direction it validates—"Voice AI replacing traditional forms"—is worth watching. Startups like TheySaid and Voiceform are the more trackable targets in this space.

Conclusion

In a nutshell: Kollect Voice Agent is an open-source experiment with excellent technical choices but is in its infancy. Its value lies not in the product itself, but in proving that the "Voice AI replacing forms" path is viable and low-cost.

Resource Links

2026-02-20 | Trend-Tracker v7.3