Frigade
Frigade is an AI-native product adoption platform for B2B SaaS teams. Its embedded assistant learns your product by actually using it, then answers questions, walks users through the live screen, or completes the action for them.
What is Frigade?
Frigade is an AI-native product adoption platform built by Frigade Inc. in San Francisco, by engineers who came from LinkedIn, Discord and Digit. Its pitch departs from tools that read your help center: Frigade's agents use your product the way real users do, walk the current flows, document what they find, and ingest whatever knowledge base you already have. Two products sit side by side. Assistant is the autonomous AI agent. Engage is a library of deterministic React components (Checklist, Tour, Announcement, Banner, Survey, Tooltip) driven no-code from the dashboard once the SDK is in place, and shipped with a Claude skill so a coding tool can write the integration against those components and open the pull request. Assistant answers in three registers. Answer replies from the product as it stands today. Show highlights the next click and runs the guided path on the live screen. Do executes the action itself through Skills, calling the actions your own team has already built. Because the agent re-runs against your product on a schedule, an interface change does not silently break anything: there are no flows to maintain, no prompts to tune and no manual element tagging. Around the assistant sit the pieces that make it a platform. Suggestions lets CS, PMM and product teams push in-app nudges from the dashboard, while a magic link is a shareable URL that opens the assistant on the right page with the right prompt. Theming inherits the customer's design system, with full white label as a paid add-on. Insights reports what users ask, where the agent helps, where it hands off, and flags spikes in questions. Governance is deliberately open: every conversation is logged to the dashboard, to Slack or through the API, and anyone can rate an answer and write the behavior they expected in plain words, applied from the next conversation onward. Installation is advertised at 15 minutes: train on staging, then drop a script tag or NPM package into production, with React, vanilla JS, modern SPAs and server-rendered apps supported. The claimed outcomes are fewer tickets, faster activation and more conversions. Named references include Vercel, Morning Consult, Merge, Valley, Hotplate, Productboard and Sanity. Frigade went through Y Combinator in Winter 2023, and Frigade AI launched on March 25, 2025.
What it does
- Answer product questions inside the app, grounded in how the product actually behaves today
- Show the way: highlight the next click and run the guided path on the live screen, across pages and modals
- Do it for the user: execute the action itself through Skills
- Fill in forms, pre-populating what it reads from the session, then confirm before submitting
- Navigate the interface: open the right pages, scroll and click
- Escalate to a human with the full session context attached
- Retrain itself on a schedule whenever the interface moves
When to use Frigade / When not to
A quick filter to help you decide if Frigade is the right fit.
When to use Frigade
- B2B SaaS companies whose interface ships faster than their help documentation can keep up
- Support teams buried under repetitive "how do I..." questions they want deflected inside the product
- Product and growth teams chasing first-hour activation and deeper feature adoption
- Enterprises that need SSO SAML 2.0, EU data residency or self-hosting alongside an in-app assistant
- Teams that want zero flow design, zero prompt maintenance and no manual element tagging
When not to use Frigade
- Individuals and small teams: the real paid entry point is 1,000 USD per month
- Anyone looking for a genuinely free tier, since the "Try Frigade / Free" card is a sales demo rather than a usable plan
- Buyers who want self-service signup, because both paid tiers run through a "Talk to us" conversation
- Anyone without a software product of their own: Frigade only exists embedded inside one
- Mid-budget teams that need EU data residency, self-hosting, the mobile SDK or webhooks, all of which are Enterprise-only
How to use Frigade
A typical end-to-end flow, from setup to results.
- Request a custom demo on your own product: send a test login and a URL, and Frigade returns a working assistant built against it, with no engineering lift on your side
- Give the agent access to your product exactly as you would a new teammate (the site shows agent@frigade.com invited and sitting Pending in a Members screen)
- Let the agent learn by using the product, working out how features chain together and where people get stuck, and let it ingest your existing knowledge base
- Train on a staging environment first, then install the SDK in production as a script tag or an NPM package (React, vanilla JS, modern SPAs and server-rendered apps are supported)
- Go live: the site claims most teams install in hours and reach their customers within a day or two
- Call the Ask API from any event to open the assistant with a prompt and context, for example frigade.ask({ prompt: "How do I export invoices?", role: user.role })
- Wire the developer surface where you need it: tool calling, event listeners, role-based dynamic permissions and redaction of sensitive fields
- Hand day-to-day operation to CS, PMM and product teams, who launch nudges, suggestions and magic links from the dashboard with no code
- Tune behavior by rating an answer and writing what the assistant should have done instead; the preference applies from the next conversation
- Sign in at app.frigade.ai for Assistant and at app.frigade.com for Engage
Pros & Cons
Pros
- The agent learns by using the product, so answers track the live interface rather than documentation that ages
- Scheduled retraining absorbs UI changes: no flows to redesign, no prompts to maintain, no manual tagging
- It goes all the way to the action through Skills and tool calls instead of stopping at an explanation
- Light installation, advertised at 15 minutes via script tag or NPM, with theming that inherits the customer's design system
- Solid security posture on paper: SOC 2 Type II, TLS 1.2+ in transit, AES-256 at rest and automatic PII scrubbing
- A published, dated subprocessor list of 21 entries updated on June 1, 2025, plus self-hosting and bring-your-own LLM keys
- Open governance and public entry pricing: full conversation logs, no-code behavior tuning by anyone on the team, and comparison pages against 11 named competitors
Cons
- An entry point of 1,000 USD per month puts the product out of reach for small teams
- No usable free tier: the "Free" card on the pricing page is a sales demo, and both paid tiers require a sales conversation rather than self-service signup
- The privacy policy and the terms of service allow Frigade to train and fine-tune its models on submitted data, which contradicts the marketing FAQ promise that data is never used for model training
- Data already incorporated into training cannot be extracted or individually deleted, and the GDPR section announces a four-year minimum retention for personal information
- Session recording is intrusive by design (mouse movements, clicks, typed text, visual captures) and opting out of training is only possible by email, with no documented self-service setting
- Many decisive capabilities are Enterprise-only: EU data residency, self-hosting, mobile SDK, webhooks, Salesforce, full white label and custom roles
- No contact page at all, everything running through modals or support@frigade.com, and no Article 27 EU representative even though the service stays reachable from the EU
Pricing & Plans
Frigade publishes no usable free plan. The pricing page shows a "Try Frigade" card at no cost, but it is a live sales demo with nothing to install rather than a working tier, and no free trial is announced anywhere on the site. The real paid entry point is the Growth plan, starting at 1,000 USD per month, which includes 5 seats, 2 AI agents and 2,500 queries, with scale discounts as volume grows. Enterprise is quoted on request and reaches up to 100 agents. Both paid tiers carry the same "Talk to us" call to action, so subscribing runs through a sales conversation rather than a self-service checkout. Billing is presented as usage-based, with predictable costs and "Cancel anytime" stated on the pricing page. Full white label is offered as a paid add-on, and custom design services are proposed to Enterprise customers.
- Free - a live demo with nothing to install and no commitment
- call to action "Get a demo"
- starting at 1
- 000 USD per month - 5 seats
- 2 AI agents
- 2
- 500 queries
- with scale discounts as you grow
- call to action "Talk to us"
- custom pricing - up to 100 agents with scale discounts
- call to action "Talk to us"
- Core Assistant
- automatic product change detection
- in-app actions (Skills)
- Generative UI answers
- proactive nudges
- brand customization
- human escalation
- tours and nudges created by prompt
- full white label
- webhooks
- PostHog
- Amplitude
- Mixpanel and Segment
- Salesforce
- custom integration support (Pylon
- Mintlify
Data, GDPR & hosting
A consolidated view of how Frigade handles your data.
GDPR overview
Section II of the privacy policy is devoted to the GDPR. Eight rights are listed: information, access, rectification, erasure, restriction, portability, objection and no automated decision-making. The legal bases invoked are contract performance, legal obligation, vital interests, legitimate interest and consent. Requests are answered within 30 days at no charge, refusal being possible for manifestly unfounded or excessive requests, with a right to complain to a supervisory authority. Breaches are notified to authorities within 72 hours and to individuals where risk is high. Frigade states it is not required to appoint an Article 37 DPO and names Christian Mathiesen, CTO, as compliance contact at support@frigade.com; no Article 27 EU representative is designated and no DPIA is performed. The company says it is headquartered in California and does not market to EU residents, while its marketing claims it is "SOC 2 Type II certified and fully GDPR compliant".
Who owns the data?
The terms of service state that Frigade Inc. does not claim ownership of User Content: it stays yours, and you alone answer for any ownership dispute over it. Submitting it, however, grants Frigade a non-exclusive, worldwide, royalty-free, sublicensable and transferable license to use, reproduce, modify, adapt, publish and create derivative works from it, expressly including AI training and product development. Aggregated Data is treated separately: the terms say Frigade "will own all Aggregated Data". Under the GDPR the company describes itself as the data controller, and elsewhere in the same policy also as a data processor, with Stripe, Inc. acting as Payments Data Controller for payments.
Reuse rights
Because Frigade does not claim ownership of User Content, customers keep their own material and need no permission to reuse it; the license runs the other way, towards Frigade. What Frigade may do with that material is broad. The privacy policy lists answering requests, running accounts, billing, marketing and newsletters, personalization, analytics and fraud prevention, and then states plainly that text inputs, uploaded content, usage patterns and feedback may be used "to train, fine-tune, and improve the performance of our artificial intelligence and machine learning systems". Amazon Web Services, Anthropic PBC, OpenAI LLC and Google LLC are named as AI subprocessors. Session recording feeds the same pipeline: mouse movements, clicks, pages visited, text typed, time per page and possible visual captures, anonymized or aggregated "wherever feasible" before model development. Note the contradiction: the commercial FAQ on the site announces "a zero-retention LLM policy so your data is never used for model training", while the binding privacy policy and terms authorize exactly that. You can ask for your data not to be used for training, but only by emailing support@frigade.com. There is no targeted advertising on personal data and no sharing of precise location without consent.
Data retention & training
Hosting summary
Frigade advertises "Cloud or self-hosted" deployment and states on its home page that hosting is available in both the EU and the United States, next to its GDPR claim. The FAQ confirms EU data residency for teams that need it, but the pricing grid ticks "EU data residency" and "Self-hosting" only at the Enterprise tier, where customers may also bring their own infrastructure and LLM keys. Beyond "EU" and "US", no individual hosting country is named, and there is no dedicated trust or security page: the information is spread across the home page, the pricing page and the subprocessor list. That list names Amazon Web Services for cloud hosting, Vercel for web hosting, Cloudflare for firewall and CDN, and Google Cloud and Microsoft Azure for model serving. Encryption is stated as TLS 1.2 or above in transit and AES-256 at rest, with automatic PII scrubbing, under a SOC 2 Type II certification monitored through Vanta. At network level the site resolves to 216.150.16.65, geolocated in the United States on AS16509 (Amazon), an anycast CDN node rather than an indication of where customer data actually lives.
Things to keep in mind
Risks and trade-offs to weigh before adopting Frigade.
- The site tells two different stories about model training: the commercial FAQ promises a zero-retention LLM policy under which your data is never used for training, while the privacy policy and the terms of service explicitly authorize Frigade to train and fine-tune its models on what you submit. Settle this in the contract, not on the marketing page
- Data already incorporated into training cannot be extracted or selectively deleted, and objecting to training is only possible by emailing support@frigade.com, with no self-service setting documented
- Personal information is kept for a minimum of four years according to the GDPR section, which is longer than many buyers assume when they sign
- Session recording is granular (mouse movements, clicks, pages visited, typed text, possible visual captures), and you remain accountable to your own end users for whatever is captured on your screens
- Frigade describes itself as both data controller and data processor within the same policy, which makes the allocation of responsibility harder to reason about
- No Article 27 EU representative is appointed even though the company acknowledges it does not block EU visitors, and it states that, given its size, it considers itself outside the scope of US state privacy laws, California included
- EU data residency, self-hosting and a custom DPA are Enterprise-only, so a Growth contract does not carry the guarantees quoted in the security narrative; note also that the site names no excluded country or territory anywhere
Setup & Integrations
Technical difficulty
Low for a product of this kind, but not zero. The site advertises a 15-minute install: train on staging, then drop the SDK into production as a script tag or an NPM package, with React, vanilla JS, modern SPAs and server-rendered apps supported; most teams are said to go live within a day or two. The genuinely technical work is granting the agent product access, placing the snippet and, optionally, wiring tool calls and role-based permissions. Everything after that is no-code from the dashboard: no flow design, no prompt tuning, no manual tagging. Enterprise deployments are assisted by Frigade's team.
Deployment
Integrations
Behind Frigade
Fundraising
Social
Resources
All the official URLs gathered for verification and reference.
Alternatives
Tools that compete with or complement Frigade.
Frequently asked questions
How does Frigade learn a product?
What happens when the product changes?
How much does Frigade cost?
How long does implementation take?
Does it work with existing support tools?
What usage data do you get back?
How are security and privacy handled?
Is customer data used to train the models?
Can Frigade be self-hosted?
Is there a minimum age?
Should you pick Frigade?
Frigade's central bet is easy to state and hard to execute: close the gap between what the documentation says and what the product actually does, by letting the agent learn through use rather than through reading. That is a genuine difference from the static-flow onboarding tools it lines up against - Appcues, Pendo, WalkMe, Whatfix, Userpilot - where someone must design, tag and maintain every path. The maturity signals are there for a company selling upmarket: SOC 2 Type II, a published and dated subprocessor list, SSO SAML 2.0, EU data residency, self-hosting and bring-your-own LLM keys. Installation stays light on paper: a snippet, a staging run, then day-to-day control handed to non-technical teams. Two reservations deserve weight before signing. The first is commercial: an entry point of 1,000 USD per month, with no self-service path and no usable free tier, closes the door on small teams, and several of the features that carry the security argument - EU residency, self-hosting, a custom DPA - sit behind the Enterprise tier. The second is legal: the marketing FAQ promises that data is never used for model training, while the privacy policy and the terms of service explicitly authorize Frigade to train and fine-tune its models on submitted content, with opt-out only by email and no way to remove data already absorbed. That contradiction is not a detail, and it should be settled contractually rather than by reading the website. Frigade is also young: a 3 M USD seed in 2023 and an AI product launched in March 2025, so roadmap risk belongs in the evaluation alongside the demo. For a B2B SaaS company with a fast-moving interface, a support queue full of "how do I..." tickets and a budget that clears the entry price, it is a serious candidate.
- Choosing a selection results in a full page refresh.
- Opens in a new window.