Live friction detection
Repeated retries, vague prompts, abandoned tasks, angry replies. Flagged the moment they happen.
Your agent should get smarter every session
FirstFlow reads every session, finds where users get lost, guides them back in real time. No deploy. No PR. The fix happens inside the conversation
[Runtime for self-improving AI agents]
Get startedYour agent should get smarter every session
FirstFlow reads every session, finds where users get lost, guides them back in real time. No deploy. No PR. The fix happens inside the conversation
[Runtime for self-improving AI agents]
Get started[]
The blank box
No one knows what to type at beginning of the conversation
Your Agent
Acme agent
Chats
User
AAWelcomeI'm here to help. What do you need?
Ask anything…
01 — This is your onboarding, currently
OpenAI
Anthropic
LangChain
Works with any LLM Stack
[]
Signals.
Behavioral detectors that flag what's happening in your sessions — with the exact quote that triggered them.
Churn signal
User threatening to cancel or asking to export data
Confusion points
User rephrasing the same request, going in circles
Hidden feature requests
User asking for something the agent can't do
Agent failures
Hallucinated facts, contradicted itself mid-conversation
Signals › Signals
Signals
Behavioral detectors that auto-flag user & agent patterns — with the exact quotes that triggered them.
Filter...
Churn signal3 matches
User expressing frustration with the product, threatening to cancel or downgrade, or asking how to export their data.
Conversations matching churn signal
- 2m ago
I need to export all my conversations before I cancel
sess_a8f3k2m9
- 18m ago
how do I download everything and close my account?
sess_p22n8w4x
- 1h ago
this isn't worth $49/mo — I'm done after this month
sess_k91d3v7q
What Firstflow does
Activation that adapts as fast as you ship.
One layer, inside the conversation. Your onboarding shouldn’t break when your product does Firstflow keeps it running.
In-chat rescue
Your agent steps in when a user is stuck: explains, shows an example, or just does the task for them.
Proof of every save
How many users got rescued, which guides did it, how much faster they hit their first win.
Retention tracking
See which first sessions turn into habits, and which save made the difference.
In-chat onboarding
Checklists and first-run guidance, delivered by your agent. No popups.
Surveys & announcements
Ask questions and ship updates inside the conversation users already trust.
Ready?
Get started[One SDK, every experience]
Every in-chat experience, from one layer.
Your Agent
Acme agent
Chats
User
AAWelcomeAsk me anything I'm here to help you get things done.
Type a message…
👋Your setup, tailored
Detected · Developer
1 of 3 complete33%
- Install the SDK
- Wrap your LLM call
- Publish your first flow
SkipContinue setup
Powered by Firstflow
Integrations
Drop it in with one line
Works with Next.js, React, Vue, NestJS, OpenAI, and Anthropic. Install the SDK on the frontend, wrap your LLM on the backend, and every call streams into Analytics & Sessions.
Install the SDK
Add the Firstflow client to your project.
Next.js
npm install @firstflow/nextjsapp/chat/page.tsx
1"use client";2import { useRef } from "react";3import { FirstflowProvider, FirstflowWidget } from "@firstflow/nextjs";45export default function ChatPage({ user, sessionId }) {6const composerRef = useRef(null);78return (9<FirstflowProvider10publishableKey={process.env.NEXT_PUBLIC_FIRSTFLOW_PUBLISHABLE_KEY!}11agentId="agt_8Kd2xQ"12conversationId={sessionId}13user={{ id: user.id, traits: { plan: user.plan } }}14>15<FirstflowWidget composerRef={composerRef} />16<Composer ref={composerRef} />17</FirstflowProvider>18);19}
Open source
Self-host dashboard, API, and SDKs every conversation in your database. Or use FirstFlow Cloud. Same code, no lock-in.
One import, and your agent starts onboarding, measuring, and retaining users today.
FAQ
FirstFlow is a runtime that reads every AI agent conversation in real time, flags where users get stuck or where the agent fails, and guides the user back inside the same chat. No popups, no PR, no waiting for the next deploy. It also surfaces signals: churn risk, confusion loops, hallucinations, hidden feature requests, each tied to the exact quote that triggered it.
Traditional tools overlay popups, modals, and tooltips on top of a UI, built for pages and buttons. AI agents don't have those. FirstFlow doesn't overlay anything: it reads the conversation itself, in real time, and intervenes inside the chat when a user stalls, gets confused, or misses something the agent could've done. The fix happens in the same message thread, not in a separate onboarding layer.
Teams shipping a chat-based AI product, a support agent, a coding assistant, a sales copilot, or any LLM-powered product where users type instead of click. If some of your users quit before getting value, FirstFlow shows you exactly where and why, and acts on it inside the session.
Every conversation runs through a set of behavioral detectors: churn signals, confusion loops, hallucinated answers, competitor mentions, hidden feature requests, and more. Each detection comes with the exact quote that triggered it, so you're reading the moment it happened, not a sentiment score.
Most teams are live in under an hour. Install the frontend SDK, wrap your LLM call on the backend, and every session starts streaming into FirstFlow. No rearchitecture, no separate flow builder to learn first.
Every session is read for topic, intent, friction, and the specific signals it triggers. You see where users stall, what they ask for that your agent can't do, and where the agent itself fails, not just that activation dropped, but the exact turn where it happened.
Yes. FirstFlow is open source. Run the whole platform, dashboard, API, and SDKs, on your own infrastructure with your own database, or use FirstFlow Cloud and skip the ops. Same SDK either way, no vendor lock-in.