Flutch
Flutch is an AI agent that handles the first conversation with every inbound lead across web, Instagram, WhatsApp and email, answering questions and qualifying against your own rules before handing sales a briefed opportunity.
What is Flutch?
Flutch is a lead qualification system published by SAB CORE LLC, a Delaware company. Its premise is simple: an AI agent takes the first conversation with every inbound enquiry, and a salesperson steps in only once the lead is qualified and the context has been written down. The agent works on the website, Instagram, WhatsApp and email; the pricing page adds phone, SMS, social DMs and referrals, all captured in a single queue.
The qualification rules are yours rather than a generic script. Need, service area, budget, timing, project detail and decision authority can each be checked, alongside any criterion your team already applies. Leads that do not qualify still get a useful answer but never interrupt the sales floor, and you decide whether they are saved, routed elsewhere or kept for later. A rep can take over at any point: Flutch alerts the right person and passes on the answers and context gathered so far.
There are two ways in. The Flutch team can deploy a configured system on your real leads in seven to ten business days, or you can build on the stack yourself. That stack has four parts. Flutch SDK is a TypeScript library that turns a LangGraph workflow into a production microservice, with REST endpoints, SSE streaming, Prometheus metrics, OpenAPI docs and graph versioning. Flutch OSS is a self-hosted runtime with a pgvector knowledge base, an admin UI and channel connectors. Both are open source on GitHub. Control Center is the SaaS console for the agent lifecycle: configuration versioning with instant rollback, A/B tests on live traffic, a knowledge hub, thread management, cost control per agent and conversation, acceptance testing, role-based access, analytics and health monitoring. Flutch Gateway is a unified LLM proxy, one API key across OpenAI, Anthropic, Google, Mistral, Cohere and Azure, with fallback chains and no markup on model costs.
Industry packs exist for home services, law firms, roofing and solar, HVAC and med-spas. Note that the performance figures on the product pages are interface mock-ups rather than published measurements.
What it does
- Answers every inbound lead immediately, on whichever channel it arrived on
- Qualifies against your own criteria: need, service area, budget, timing, project detail, decision authority
- Answers product and service questions from your approved services, pricing, FAQs and policies, with sources
- Hands sales a lead card carrying source, priority and the full conversation summary
- Chases quiet leads on its own on day 2, day 5 and day 10
- Syncs the lead and the entire transcript both ways with your CRM
- Books site visits, sends quotes with a Stripe pay-link and requests a Google review after the job
When to use Flutch / When not to
A quick filter to help you decide if Flutch is the right fit.
When to use Flutch
- Small and mid-sized service businesses whose reps spend the day filtering enquiries instead of selling
- Roofing, solar, HVAC and home-services companies, which get a dedicated industry pack and per-service-line qualification flows
- Law firms, med-spas and clinics that need consistent intake before a fee-earner picks up the file
- Sales and revenue leaders who want every lead answered within seconds without hiring another rep
- Engineering teams that would rather build or self-host their own agents on an open-source SDK and runtime
When not to use Flutch
- Solo operators and very small teams: the real entry ticket is a mandatory USD 990 build plus USD 99 a month, with AI tokens on top
- Anyone expecting a native mobile app, since the terms describe responsive web only, on Safari iOS 14+ and Chrome Android 10+
- Non-English operations: no interface or processing language is declared anywhere on the site
- Organisations that require EU data residency on a hosted plan, because the SaaS keeps data in the United States
- Buyers who need verifiable references before signing: not one named customer, case study or usage figure is published
How to use Flutch
A typical end-to-end flow, from setup to results.
- Paste your website URL on the homepage and let Flutch read your pages and assemble a first knowledge base
- Review what the agent says, then claim the account: registration comes after the build, not before
- For an assisted rollout, start with discovery, which covers stakeholder interviews, workflow review and CRM analysis and produces an Implementation Blueprint
- Hand over your sales scripts, objection-handling material and pricing documentation so the agent can be configured on them
- Define the qualification logic, the lead channels, the follow-up sequences and the routing rules
- Connect your CRM, whether Zoho, Jobber, HubSpot, GoHighLevel, Pipedrive, ServiceTitan or Salesforce
- Run scenario testing and QA review on the scripts before anything reaches a real lead
- Go live, take the two-hour training session and the user documentation, then use the thirty days of included support
- Compose your commercial package on the pricing calculator: industry pack, number of service lines, hosted or self-hosted delivery, support tier
- Developers can bypass all of this: npm install -g flutch, then flutch graph create, register and deploy, with Python and TypeScript SDKs documented
Pros & Cons
Pros
- Pricing is fully public and calculable on screen, upper tiers included, which is unusual in this segment
- AI tokens are re-billed at provider cost with no markup, against a published per-model price table
- You can leave the hosted plan: migration to the self-hosted build is a fixed USD 2,500 one-time, with code and data on your own infrastructure
- Two of the four components are open source, with public GitHub repositories
- Substantial public technical documentation: 29 articles, Python and TypeScript SDKs, a CLI and an agent API
- Legal and security documentation is unusually complete for a company this size, with a trust centre, sub-processors named by region, SOC 2 Type II reported, PCI-DSS SAQ A, SCCs and the UK addendum
- No training on customer data by default, since it takes an explicit opt-in that can be withdrawn
Cons
- A very young and thinly documented publisher: domain registered 27 May 2025, first Wayback capture 20 September 2025, no named director, no social account, no identified funding
- No verifiable customer proof, with no client logo, no quantified case study and no named testimonial; the five demos linked from the homepage are fictional businesses hosted on AWS Amplify
- The latency, error-rate and A/B figures shown on the product pages are interface mock-ups rather than published measurements
- The terms of use govern an agent marketplace with a free tier and Premium, Pro and Max plans that bears no relation to the pricing page, so the contract does not describe the service being sold
- Several official links are dead: the declared og:image, the footer link to the terms, flutch.ai/privacy, flutch.ai/cookie-policy, status.flutch.ai, and a mailto:undefined closing the privacy policy
- Leftovers from an earlier brand and positioning remain visible, the legal page still being titled Agentech while /pricing and /docs still call the product an AI observability platform
- No declared interface language, no native mobile app, and ISO 27001 still in progress with an audit scheduled for Q4 2025 and no update since
Pricing & Plans
Flutch does not publish a free plan on its pricing page, and the site does not clearly confirm a free trial: the terms of use mention a free tier that the commercial offer contradicts. Access begins with a mandatory one-time core build of USD 990, to which a support tier must be added. The lowest of these is Self-service at USD 99 per month. An industry pack costs USD 600 per service line, and self-hosted delivery adds USD 2,500 one-time. AI tokens are billed separately at provider cost on every tier.
- multi-channel capture and qualification
- conversational agent
- lead card
- day 2/5/10 follow-up
- instant alerting
- bi-directional CRM sync
- knowledge base and live funnel dashboard
- online quote with Stripe pay-link
- booking with SMS and email reminders
- automatic Google review request and a per-service-line script
- SaaS delivery
- included in the support tier
- with AI tokens billed on top
- code and data on your own infrastructure
- with AI tokens billed on top
- hosting
- isolated environment
- daily backups
- industry pack pinned as built
- all changes on your side
- ad-hoc help at USD 500 per hour
- automatic industry-pack updates
- 8 hours of hands-on work monthly
- a 12-hour incident SLA and 3 business days on new tasks
- overage at USD 300 per hour
- named engineer with 2-hour response
- 15 hours monthly
- overage at USD 200 per hour
- funnel analytics on your data and a quarterly strategic review
- dedicated engineer and analyst 24/7 with 30-minute response on critical incidents
- 40 hours monthly
- overage at USD 150 per hour
- continuous improvement loop
- segmented analytics
- monthly audit and quarterly business reviews
Data, GDPR & hosting
A consolidated view of how Flutch handles your data.
GDPR overview
GDPR compliance is claimed in writing: the trust centre lists GDPR and UK GDPR as compliant for data processing, citing a DPA template and a published sub-processor list as evidence. The privacy policy grants access, rectification, erasure, restriction, portability, objection and withdrawal of consent, with a thirty-day response window and requests sent to privacy@flutch.ai. Transfers of EEA and UK data to the United States rely on standard contractual clauses and the UK addendum. Complaints are directed to the UK Information Commissioner's Office or a local authority, and Californian residents are given separate CCPA and CPRA rights. Two gaps deserve naming: no Article 27 EU representative is designated anywhere on the site, and no data protection officer is named, privacy@flutch.ai being the only contact point. The privacy policy is dated 28 May 2025, the trust centre 27 November 2025.
Who owns the data?
Under the terms of use, SAB CORE LLC owns all rights in the platform itself, while you keep whatever rights you already hold in the prompts and content you submit. Rights in generated outputs stay with you, subject to applicable law and to the upstream conditions of the model providers, OpenAI and Anthropic. Submitting content grants the company a non-exclusive, worldwide licence, but strictly to host, display and use that content to operate, secure and improve the platform. SAB CORE LLC is the controller for personal data and is reached at privacy@flutch.ai. On termination you may export your data for thirty days, after which it is deleted.
Reuse rights
You may reuse what you generate without asking permission: the terms state that, subject to applicable law and to the model providers' own conditions, rights in outputs you lawfully generate belong to you, and nothing restricts commercial use of them. The company's own reuse is deliberately narrower. Your content and outputs are not used for commercial purposes unrelated to your use of the platform, and not used to train Flutch's models or third-party models, unless you switch that on yourself through a setting in the account dashboard, a consent you can withdraw at any time. Special-category data caught in a prompt is used only to produce the answer, is kept out of training and analytics, and is deleted within ninety days. Personal data is not sold.
Data retention & training
Hosting summary
On the hosted plans, data stays in the United States. The trust centre names AWS in us-east and us-west for primary storage and MongoDB Atlas for the managed database, with model requests processed in the US by the AI providers. The published sub-processor list gives a region for each vendor: AWS, MongoDB Atlas, Stripe, OpenAI, Anthropic, PostHog Cloud US, SendGrid and Ketch, all in the USA. The terms add Intercom for support and the privacy policy adds Google among the model providers. Changes to that list are announced thirty days in advance. Data is encrypted with TLS 1.2 or later in transit and AES-256-GCM at rest. Network controls include VPC segmentation, private subnets and a CloudFront WAF with OWASP rules, while access relies on SSO/SAML, enforced MFA and least-privilege IAM roles, with secrets held in AWS Secrets Manager. Dependency scanning runs daily, external penetration testing quarterly, and critical CVEs carry a 24-hour SLA. Organisations that need data outside the United States have one route: Flutch OSS, where everything runs on their own infrastructure.
Things to keep in mind
Risks and trade-offs to weigh before adopting Flutch.
- The bill is elastic exactly where it matters: AI tokens are never included and are metered per conversation, so a good month for lead volume is also an unbudgeted month
- The contract you sign may not describe the service you bought, since the terms of use govern an agent marketplace with a free tier and Premium, Pro and Max plans rather than the package on the pricing page
- Letting an agent hold every first conversation slowly erodes a team's feel for objections, wording and buying signals, and the reps who learn least from it are the juniors who need that practice most
- Qualification rules encode judgement: a badly drawn budget threshold or service-area filter silently discards real customers, and nobody looking at the funnel dashboard sees the people who were turned away
- The demos linked from the homepage are fictional businesses, and the latency, error-rate and A/B numbers on the product pages are interface mock-ups, both of which are easy to read as results
- Full transcripts flow into your CRM, so whatever a prospect volunteers about their health, finances or household ends up in a system built for selling rather than for sensitive data
- Trust signals here need checking rather than believing: the SOC 2 report is released only under NDA, ISO 27001 is still in progress, the announced status page does not resolve, and several official links are dead
Setup & Integrations
Technical difficulty
It depends entirely on the route. The assisted rollout asks nothing technical of you: the Flutch team wires the channels, the CRM and the knowledge base while you supply the business knowledge, over seven to ten business days, ending with a two-hour handover and thirty days of support. Day-to-day changes are then no-code, through the web interface. Building it yourself is a different job, requiring Node and TypeScript, a CLI, LangGraph concepts, and Docker with PostgreSQL and pgvector for self-hosting. Note that on the Self-service tier every change stays on your side.
Deployment
Integrations
Behind Flutch
Resources
All the official URLs gathered for verification and reference.
Frequently asked questions
What counts as a qualified lead?
Which channels can Flutch handle?
What happens to leads that are not qualified?
Can a salesperson take over the conversation?
What does it cost to get started?
How long does it take to go live?
Can I host it myself?
Is my data used to train AI models?
Where is the data hosted, and what certifications exist?
Should you pick Flutch?
Flutch answers a real and unglamorous problem: the first-line filtering of inbound leads, which quietly consumes the day of people hired to sell. On that problem it is well equipped, with qualification rules that follow your process rather than a generic script, industry packs for home services, law firms, roofing and solar, HVAC and med-spas, and a handover that gives the rep the whole conversation rather than a name and a phone number.
Two things set it apart from the crowd. Pricing is genuinely public and calculable on screen, including the upper tiers, and model tokens are re-billed at provider cost with no markup. And the exit is priced: for a fixed USD 2,500 you move to the open-source build and run code and data on your own infrastructure. Add two open-source components, a documented SDK and a trust centre that names its sub-processors by region, and the transparency is well above the segment average.
The reservations are about the publisher, not the product. The domain was registered in May 2025 and first archived in September 2025; no director is named, no social account exists, no funding is identifiable, and not a single customer reference is published. The site still carries at least two earlier positionings, with terms of use that govern an agent marketplace rather than the package being sold, page titles inherited from an AI observability product, and several dead official links.
The natural buyer is a service business with real inbound volume in one of the packaged verticals, or an engineering team drawn to the open-source runtime. Before committing, ask for the SOC 2 report, ask which contract actually governs the service, and ask to speak to a live customer.
- Choosing a selection results in a full page refresh.
- Opens in a new window.