Why We Built FeedbackFlow
Every product team has the same problem: users report bugs in Slack, feature requests arrive via email, and support tickets describe issues with "it's broken" and nothing else. Meanwhile, developers play detective, trying to reproduce problems from vague descriptions while real work sits idle.
We know this pain intimately. Running multiple products simultaneously means feedback comes from everywhere. Notion comments, GitHub issues, Twitter DMs, Discord messages. Each requires context switching, manual triage, and the tedious work of turning "the button doesn't work" into an actionable ticket with steps to reproduce.
The existing tools didn't fit our workflow. Hotjar captures behavior but not intent. UserVoice is built for enterprise feature voting, not fast-moving product development. Zendesk is a support system masquerading as feedback collection. None of them understood what we actually needed: feedback that flows directly into AI-powered development.
So we built FeedbackFlow. A lightweight widget that captures rich context—screenshots with annotations, screen recordings with voice narration—and a dashboard that uses AI to turn raw feedback into complete, actionable tickets. Export to Linear or Notion if you're using traditional workflows. Or export to prd.json if you're building with AI agents.
What is FeedbackFlow?
FeedbackFlow is a feedback collection system with two parts:
The Widget: A lightweight JavaScript embed (under 50KB) that adds a feedback button to any website. Users can capture screenshots with annotation tools or record their screen with audio narration. The widget works offline—submissions queue and retry automatically.
The Dashboard: A central command center for processing feedback. AI analyzes each submission to suggest categorization, priority, and potential solutions. Draft complete tickets with one click. Export to Linear, Notion, or JSON. Assign to team members and track status through resolution.
The system is designed around a core belief: feedback is only valuable if it becomes action. Raw user reports sitting in a database help no one. FeedbackFlow's job is to reduce the time between "user reports problem" and "developer fixes problem" to near zero.
Built for AI Development Workflows
Here's what makes FeedbackFlow different from every other feedback tool: prd.json export.
If you're using AI coding agents—Claude, GPT, or custom agents—you know the bottleneck isn't the AI. It's getting requirements into a format the AI understands. Most teams waste hours reformatting feedback into structured specifications that AI can execute.
FeedbackFlow's JSON export generates prd.json format directly:
{
"id": "FF-001",
"title": "User reported bug in checkout flow",
"acceptanceCriteria": [
"Payment form validates card before submission",
"Error message displays for invalid card numbers",
"User can retry payment without refreshing"
],
"priority": 1,
"notes": "User recorded screen showing card validation failing silently"
}
This isn't just JSON. It's the exact format AI development pipelines expect. Feed it directly to your agent. The AI has context, acceptance criteria, and priority—everything needed to implement the fix.
We use this ourselves. Raw feedback comes in through the widget. AI drafts the ticket. We review, adjust priority, and export to our prd.json. Within minutes, AI agents are working on the fix while we move to the next item. The human review loop takes 30 seconds per item instead of 30 minutes.
How FeedbackFlow Works
Capture
Embed the widget on your site with a single script tag:
<script
src="https://cdn.feedbackflow.cc/widget.js"
data-widget-key="wk_xxxxx"
></script>
Users click the feedback button to open a capture modal. They choose screenshot or recording, add annotations or narration, fill in a quick form with title and description, and submit. The widget handles compression, offline queuing, and retry logic automatically.
Process
Submissions appear in your dashboard inbox. Each item shows a thumbnail, type badge, metadata (browser, OS, URL), and submission time. Click to open the detail view with full screenshot or video playback.
If you've configured an AI API key, FeedbackFlow analyzes the feedback automatically:
- Suggests type (bug vs feature request)
- Recommends priority based on content
- Proposes tags from your taxonomy
- Offers solution suggestions
The real power is ticket drafting. Click "Draft Ticket" and AI generates a complete specification: clear title, detailed description, reproduction steps, expected vs actual behavior, and acceptance criteria. Edit if needed, then export.
Export
Three primary export destinations:
Linear: Creates issues with labels, attaches media, links back to FeedbackFlow. Optional status sync keeps everything updated.
Notion: Creates pages in your selected database, maps properties, embeds screenshots directly in the page body.
JSON (prd.json): The format AI coding agents understand. Export single items or bulk export your entire backlog. Download as file or copy to clipboard.
Webhooks and a REST API handle custom integrations. Automation rules can export automatically—all critical bugs to Linear, all feature requests to Notion, everything to your data warehouse.
Why Open Source?
We're open sourcing FeedbackFlow because the feedback tools we needed didn't exist, and we believe in building in public.
Self-hosting matters. Some teams can't send user data to external services. Regulated industries, enterprise security requirements, or simple preference for data ownership. FeedbackFlow self-hosts completely—deploy your own instance with Docker, bring your own database and storage, and own every byte of feedback data.
No artificial limits. Self-hosted FeedbackFlow has no seat limits, no feedback caps, no feature gates. Use every capability on your own infrastructure without paying per-seat pricing.
Community improvements. The best feedback tool is one shaped by people using it daily. Open source means bug fixes merge faster, features get built that we'd never prioritize, and edge cases get handled by people who actually experience them.
Trust through transparency. You can audit every line of code that handles your user's feedback. No wondering what the feedback tool is doing with your data. No surprise telemetry or tracking.
We still offer a hosted version for teams who want managed infrastructure. Free tier for individuals, Pro tier for teams with unlimited feedback. But the self-host option exists for anyone who wants it, and it's the exact same codebase.
The Feedback-to-Fix Pipeline
FeedbackFlow fits into a larger vision: autonomous product development where AI handles the tedious work while humans make decisions.
The pipeline looks like this:
- User reports issue via FeedbackFlow widget
- AI triages automatically—categorizes, prioritizes, suggests fixes
- Human reviews draft ticket, adjusts if needed (30 seconds)
- Export to prd.json for AI development pipeline
- AI agent implements the fix based on structured requirements
- Human reviews the code, merges if good
- User notified through FeedbackFlow status page
Each step involves human judgment where it matters—reviewing AI output, making priority decisions, approving code changes. But the mechanical work of turning vague feedback into actionable specifications happens automatically.
This is the future of product development. Not replacing humans with AI, but removing the drudgery so humans can focus on decisions that matter. FeedbackFlow is one piece of that puzzle: the piece that turns user feedback into structured requirements AI can execute.
Key Features
Lightweight Widget: Under 50KB, async loading, no dependencies. Screenshot capture with annotation tools. Screen recording with audio up to 2 minutes. Offline queue with automatic retry.
AI Processing: Auto-categorization, priority suggestions, solution recommendations, and complete ticket drafting. Bring your own OpenAI or Anthropic API key—your data never touches our servers.
Multi-Destination Export: Linear, Notion, JSON (prd.json), webhooks, REST API. Automation rules for auto-export based on type, priority, or tags.
Team Collaboration: Assign feedback to team members. Comment threads for discussion. Activity log for audit trails. Real-time updates via Convex.
Search & Filter: Full-text search across titles, descriptions, comments, and tags. Filter by type, status, priority. Sort by date or priority.
Self-Host Option: Docker deployment, configurable storage (S3/R2/GCS), no artificial limits. Same codebase as hosted version.
Who Should Use FeedbackFlow
Teams using AI coding agents: If you're building with Claude Code, GPT, or custom agents, the prd.json export eliminates the bottleneck of formatting requirements.
Indie hackers and small teams: The free tier and self-host option mean you can collect professional-grade feedback without enterprise pricing.
Product teams drowning in feedback channels: Consolidate feedback from widgets instead of scattered Slack messages, emails, and support tickets.
Teams with compliance requirements: Self-host for complete data ownership and auditability.
Open source maintainers: Collect bug reports and feature requests with rich context instead of incomplete GitHub issues.
Getting Started
FeedbackFlow is currently in development with core features complete. The widget captures screenshots and recordings. The dashboard handles triage and team collaboration. Search, filtering, and comments work.
Coming soon: AI processing features (auto-categorization, ticket drafting), Linear and Notion integrations, and the prd.json export that makes FeedbackFlow unique.
Self-hosting documentation will be available at launch. For teams who want managed infrastructure, the hosted version will offer a free tier for individual users and Pro pricing for teams.
Watch the GitHub repo for updates. We're building in public because that's how we believe tools should be built—with input from the people who'll actually use them.
Related Reading
- ValidateFirst: AI-Powered Startup Validation - Validate ideas before building
- FounderFlow: Project Management for Founders - Track your product journey
- Psyche: Psychedelic Music Visualizer - See how we build creative consumer tools
The Feedback Loop Closes Here
FeedbackFlow exists because we got tired of the same broken workflow: user reports bug → developer plays detective → hours wasted → bug eventually fixed. That loop is too slow for modern product development, especially when AI can handle the translation from "it's broken" to structured requirements.
If you're building products and collecting feedback, you need a system that treats feedback as the starting point of development, not an afterthought to manage. FeedbackFlow is that system. Open source, AI-native, and designed for the way modern teams actually work.
The gap between user feedback and shipped fix should be hours, not weeks. FeedbackFlow makes that possible.