Myriade
Myriade is an AI-native data platform that connects read-only to your warehouse, then uses autonomous agents to document tables, validate metrics and flag anomalies, so natural-language analytics answer correctly without a semantic layer built upfront.
What is Myriade?
Myriade is an AI-native data intelligence platform built on a simple observation: most companies fail at AI-on-their-data not because the models are weak but because the data underneath is a mess. Incomplete catalogs, contradictory metrics, undocumented tables and a decade of decisions nobody remembers are the normal state of an enterprise warehouse, and a generic assistant dropped on top of that will produce numbers that look plausible and are wrong.
Rather than demanding a cleanup first, Myriade plugs into the warehouse as it is, read-only, and puts AI agents to work on the reverse engineering. They sample data, run statistical analysis, read whatever metadata already exists in dbt or a prior catalog, infer table relationships from the JOINs your team actually executes, detect anomalies and propose metric definitions. The vendor frames this on an hourly timeline: connection at hour zero, autonomous discovery by hour four, human validation by hour twenty-four, trustworthy analytics by hour forty-eight. Around 90% of tables are documented without supervision, at a claimed five minutes per fifty tables.
The part that distinguishes it from a chatbot is the loop back to people. When a definition is genuinely ambiguous — the vendor's own example is whether a price column is in dollars or euros — the agent asks the data team instead of guessing. Humans approve, correct, or reject; the volume is handled for them.
Five modules share one layer: Documentation, Context, Analytics with cited natural-language querying, Quality with continuous freshness and consistency checks, and Transformations that generate dbt models and SQL optimisations. A FinOps agent audits compute waste, with the vendor claiming 27-40% savings in a first quarter and guaranteeing the sprint or waiving payment.
Myriade is deliberate about what it is not: not a BI tool, not a replacement for your warehouse, not a replacement for your catalog — which it completes and syncs back to — not a one-off audit, and not a pipeline migration tool. It complements dbt rather than displacing it, and the customer's dbt project stays the source of truth.
What it does
- Document tables and columns across an undocumented warehouse automatically, then keep those descriptions current
- Answer business questions in natural language, showing the SQL executed and the tables cited
- Detect data quality anomalies continuously and raise them before they reach a dashboard
- Map relationships and lineage between tables from the JOINs actually run in query history
- Generate dbt models, tests and fix suggestions from an intent described in plain language
- Audit warehouse compute spend and quantify where money is being wasted
- Serve governed data to Claude, Cursor and other LLM clients through a built-in MCP server
When to use Myriade / When not to
A quick filter to help you decide if Myriade is the right fit.
When to use Myriade
- Data engineers and analytics engineers buried under undocumented tables, ad-hoc requests and legacy pipelines nobody remembers building
- Chief Data Officers and Heads of Data who need AI on company data to actually be trustworthy before opening it to the business
- Data stewards and data quality analysts who own a catalog that is technically alive but practically empty
- Multi-channel retail and e-commerce organisations reconciling stores, marketplaces, CRM and logistics into one set of agreed numbers
- Finance, operations and supply chain teams that want their own answers without queuing behind the data team for every question
When not to use Myriade
- Anyone shopping for a BI or dashboarding tool: the vendor states plainly that Myriade is not one and expects to sit alongside Metabase, Looker, Tableau or Power BI
- Teams with no data warehouse or SQL database to connect, since the whole product starts from a read-only warehouse connection
- Organisations hoping to migrate or rewrite legacy pipelines, a use case the vendor excludes outright
- Small teams looking for a free tier or month-to-month billing: every plan carries a twelve-month commitment and a seat minimum
- Buyers whose procurement requires a signed SOC 2 or ISO 27001 certificate, which Myriade does not yet hold
How to use Myriade
A typical end-to-end flow, from setup to results.
- Create a dedicated read-only service account on your warehouse, granting access to INFORMATION_SCHEMA plus the schemas you want analysed
- Grant SELECT only: on Snowflake this means a role such as MYRIADE_READER with USAGE on the warehouse, database and schema
- Enter the service account credentials in the Myriade dashboard; the vendor budgets five to ten minutes per platform
- Let the discovery agents run: they sample data, infer table meaning, map relationships and draft descriptions unsupervised
- Review the queue of items the agents flagged as ambiguous and approve, edit or reject each proposed definition
- Ask your first business question in plain language and read the answer as four blocks: exploration log, SQL executed, results with statistical context, and quality notes on assumptions made
- Invite business users as Standard or Viewer seats once the catalog is documented and quality-checked
- Turn on the quality agents so freshness, completeness and consistency issues are raised before dashboards break
- Run the FinOps audit to surface unused tables, inefficient scheduled queries and oversized clusters
- For a self-hosted deployment, run the Docker container on your own infrastructure instead of using Myriade Cloud
Pros & Cons
Pros
- Published, itemised pricing down to the seat and the credit pack, which is rare in this segment and makes budgeting possible before a sales call
- Flat platform and seat fees rather than billing tied to warehouse compute, so the invoice does not move with usage the vendor cannot see
- Self-hosting via Docker on Business and Enterprise, with zero data retention and no training on customer data by Myriade or its model provider
- An explicit human-in-the-loop design: the agent asks when a definition is ambiguous instead of inventing one, and every query and decision is logged and auditable
- Genuinely portable output: the semantic layer, definitions and validation history export in open formats, and multiple warehouses can be queried together
- Concrete technical documentation, including the SQL needed to create a read-only role per platform, rather than marketing copy
- A 30-day pilot on the customer's own data with four workshops and a closing report, plus a FinOps audit sold on guaranteed savings
Cons
- No legal notice anywhere on the site: no full company name, legal form, registration number or postal address, so the publisher's identity cannot be verified from the site itself
- Terms and privacy policy are brief and still dated November 2024 on a site that otherwise reads 2026, and neither document mentions the GDPR despite the compliance claim
- Neither SOC 2 nor ISO 27001; the vendor answers "not yet" and cites a three-person team, which will block some enterprise procurement outright
- A three-person team behind an enterprise-grade purchase raises real questions about support capacity and continuity
- No permanent free plan and a twelve-month minimum commitment on every plan, with seat minimums of two, five and twenty, so entry cost exceeds the headline platform fee
- The site carries contradictory security claims, showing "SOC 2 Type II Ready" and "SOC 2 & HIPAA Compliant" elsewhere while the security FAQ says no certification exists
- Credit-based usage governed by an unquantified fair-use policy makes heavy consumption hard to forecast, and self-hosting is unavailable on the entry plan
Pricing & Plans
There is no permanent free plan. The lowest entry point is the Startup plan at EUR 149 per month as a platform fee, billed monthly or EUR 1,490 annually, and all plans require a twelve-month commitment with prices quoted excluding VAT. Because seats are charged separately and a minimum of two applies on Startup, the effective starting cost is the platform fee plus at least two seats, from EUR 9 per month for a Viewer to EUR 69 per month for a Power user. A 30-day pilot is offered in place of a free tier.
- minimum 2 seats
- 1 SQL source
- hosted AI
- 240 organisation credits plus 150 per Standard and 600 per Power seat
- 3
- 000 API calls/month
- standard SSO
- centralised hosting
- minimum 5 seats
- up to 4 data warehouses
- hosted AI
- 1
- 200 organisation credits plus 240 per Standard and 960 per Power seat
- 10
- 000 API calls/month
- standard SSO
- minimum 20 seats
- unlimited sources
- your choice of AI whether hosted or bring-your-own-key
- fair-use organisation credits
- unlimited API calls
- SAML and SCIM
- VPC or self-hosted deployment
- one CSM onboarding and four data team workshops per year
- Standard for querying and catalog access
- Power for full access including deep analysis and transformation agents
- Viewer for read-only at EUR 9 per month on every plan
Data, GDPR & hosting
A consolidated view of how Myriade handles your data.
GDPR overview
Myriade claims GDPR compliance in plain words on its security page, arguing that it processes minimal data, keeps it no longer than necessary, provides a Data Processing Agreement on request, and that self-hosted deployments never let data leave the customer's infrastructure. The privacy policy lists access, correction, deletion, portability and marketing opt-out rights, and the vendor advertises encryption in transit and at rest. The claim is nonetheless stronger than the paperwork behind it. The privacy policy, last updated in November 2024, never mentions the GDPR at all: no legal basis, no stated retention period, no transfer mechanism, no subprocessor list, and no Article 27 representative anywhere on the site. No DPO is named either; privacy@myriade.ai is the only privacy contact. Neither SOC 2 nor ISO 27001 has been obtained.
Who owns the data?
Your source data stays in your own warehouse; Myriade connects to it with read-only credentials and never moves it. In self-hosted deployments nothing leaves your infrastructure at all, and on Myriade Cloud only metadata and conversation history are stored. The semantic layer the agents build remains yours in practice: definitions, documentation and validation history can be exported at any time in open, portable formats. One gap is worth naming. The terms of service state that Myriade's content, features and functionality belong to Myriade, but they say nothing explicit about ownership of your data or of the agent outputs generated from it. That silence is not the same as a guarantee.
Reuse rights
The published terms grant you use of the service without setting out any licence you must request before reusing what comes out of it, and the exported semantic layer is presented as yours to take with you. On the vendor's side, use is described narrowly: account details, usage analytics, your messages and the data you connect are processed to run and improve the service, secure it and keep you informed. Personal information is not sold. It may be shared with service providers operating the platform, with authorities under legal obligation, and with commercial partners only on your explicit consent. Crucially, neither Myriade nor its default model provider trains on your data, and the AI works from table structures and samples rather than full extracts. What the documents do not do is spell out a licence over your data or the agents' output, so a negotiated agreement remains the place to settle that.
Data retention & training
Hosting summary
Myriade offers two deployment models and they differ sharply in where data lives. With Myriade Cloud, the vendor hosts the platform on Google Cloud Platform, handles maintenance and updates, and makes audit logs available on request. Self-hosted deployment, available on the Business and Enterprise plans, runs a Docker container on the customer's own infrastructure: data never leaves the customer's network, the customer controls updates and access, and full audit logs stay on the customer's systems. The entry Startup plan is limited to the vendor's centralised infrastructure. In both models the source data itself is not relocated: it remains in the customer's warehouse and Myriade connects to it read-only. Two limits are worth noting. No hosting country or region is stated beyond the mention of GCP, so no jurisdiction can be cited with confidence. And the website is served behind Cloudflare on an anycast address, which means the resolved server location says nothing about where the product is actually hosted.
Things to keep in mind
Risks and trade-offs to weigh before adopting Myriade.
- The publisher cannot be identified from its own site: no legal notice, no registered company name or legal form, no registration number, no postal address. If a dispute arises, you would not know from the website who you are contracting with
- A claimed GDPR compliance that the paperwork does not support: the privacy policy never mentions the regulation, gives no legal basis, no quantified retention period, no subprocessor list and names no Article 27 representative, and it has not been updated since November 2024
- Contradictory security signals on the same site: the security FAQ says no SOC 2 or ISO 27001 certification exists, while other pages advertise "SOC 2 Type II Ready" and "SOC 2 & HIPAA Compliant". Trust the explicit answer, and ask before signing
- Concentration risk on a three-person team: support capacity, incident response and simple business continuity are real exposures for a tool wired into a production warehouse
- Cost that is predictable in structure but not in total: a twelve-month lock-in, seat minimums, and credit consumption governed by an unquantified fair-use policy mean the bill can outrun the headline figure
- Automation bias is the human risk here. Roughly 90% of documentation arrives pre-written and confidence-scored, and a reviewer under time pressure will rubber-stamp it. The tool is built to ask when unsure, but only a team that genuinely reviews the 10% keeps that safeguard alive
- Fluent answers can erode the habit of checking. Every query and citation is exposed precisely so it can be verified; a business user who stops opening the SQL is trusting a chain they no longer inspect, and the vendor's own ROI figures of 27-40% savings or 2-3x capacity are unaudited claims
Setup & Integrations
Technical difficulty
Moderate, and squarely a job for someone with warehouse administration rights rather than a business user. You must create a dedicated read-only service account, grant it access to INFORMATION_SCHEMA and the schemas to be analysed, and enter the credentials in the dashboard; the vendor budgets five to ten minutes per platform and publishes the required SQL commands. Self-hosting adds a Docker deployment on your own infrastructure. Full pilot setup is quoted at 72 hours. End users need nothing technical: questions are asked in plain language, with no SQL required.
Deployment
Integrations
Supported languages
Behind Myriade
Fundraising
Social
Resources
All the official URLs gathered for verification and reference.
Alternatives
Tools that compete with or complement Myriade.
Frequently asked questions
Does Myriade replace my data catalog or my dbt project?
What does the AI actually see of my data?
Is my data used to train AI models?
Is Myriade SOC 2 or ISO 27001 certified?
Is there a free plan or a free trial?
How technical is the setup, and who has to do it?
How do the AI credits work?
Who is behind Myriade, and how transparent is the company?
Which platforms and tools does Myriade connect to?
How quickly should we expect results?
Should you pick Myriade?
Myriade is a substantial, unusually well-documented product wrapped in a strikingly thin corporate shell, and both halves of that sentence matter.
The product side holds up to scrutiny. Pricing is published to the level of individual seats and credit packs. The documentation contains the actual SQL you need to create a read-only role on each platform. There is a working demo, a public instance on open data, named and checkable customers including Promod, Jules, Carter Cash and Recommerce, and testimonials attributed to real people with titles and employers. The strategic argument against warehouse-native AI — keep your context layer portable, query several sources at once, pay a flat fee, and let the vendor's FinOps incentives align with yours — is coherent rather than decorative. Security posture is serious for the size of the outfit: read-only access, self-hosting, zero retention, no training on customer data, field-level masking and bring-your-own-key.
The corporate side is where caution belongs. There is no legal notice, no company address, no legal form. The terms and privacy policy are short and frozen at November 2024, and neither mentions the GDPR even though the security page claims full compliance. No SOC 2, no ISO 27001, a self-declared three-person team, and contradictory certification badges elsewhere on the same site. Every plan locks in twelve months with a seat minimum, so real entry cost exceeds the advertised fee.
The honest reading is that this suits a data team willing to accept a very young vendor in exchange for concrete automation and a predictable bill. The 30-day pilot is the right instrument for that bet: run it on your own data, and use the procurement conversation to obtain in writing what the website does not publish — corporate identity, data ownership terms and a signed DPA.
- Choosing a selection results in a full page refresh.
- Opens in a new window.