Telmai

Editor’s note: This Telmai review is based on my hands-on test of Telmai v26.5.0 on the SaaS deployment, in November 2026. I weighed pricing and packaging, setup and the first scan, monitors and ML alerting, lakehouse and file coverage, deployment and security, remediation and AI features, and how it compares with Monte Carlo, Bigeye, Anomalo, Sifflet and open-source checks.

Quick verdict: I recommend Telmai for data teams that need to monitor Iceberg, Delta and Parquet data in object storage as well as warehouse tables, especially when policy keeps row data inside their own cloud (Hybrid or VPC). The trade-off is no public price list (the one public figure is a 12-month AWS Marketplace contract at $60,000) and no self-serve signup; Soda publishes a Free plan and Team at $750 a month.

Telmai combines 11 out-of-box monitors that start learning as soon as a source is connected with a Spark engine that scans Snowflake, BigQuery, Databricks and Redshift plus CSV, JSON, Parquet, Avro, Delta and XML files in S3, Google Cloud Storage and Azure, earning my recommendation for mixed warehouse and file monitoring in this Telmai review. Hybrid mode runs scans on my own Spark engine, so no row-level data leaves my environment.

In this review, I’ll take a closer look at Telmai’s pricing, features, and deployment options, so you can see exactly whether it fits your data team or whether a warehouse-first tool like Monte Carlo is the better buy.

Key Takeaways

• Monitors start on connection: Learn what’s normal set thresholds from recent scans, so I wrote no baselines.
• Hybrid and VPC keep row data in my cloud; Telmai stores only derived metrics and metadata.
• No public pricing or self-serve signup: logins come from Telmai Support; the one public figure is a $60,000 AWS Marketplace contract.
• Coverage gaps: no Kafka or streaming connector, no native PagerDuty, and no freshness for file-based assets.
• Monthly releases: v26.5.0 (November 9, 2026) added Ask AI, incident blast radius, ServiceNow and Atlan.

Pros and Cons

Telmai Pricing: How Much Does It Cost?

Telmai has no public price list. Its terms describe tiered subscription plans with a free trial, usage-based tiers and enterprise arrangements, and the only published dollar figure is the AWS Marketplace contract.

• Telmai 60K ($60,000 a year): for teams buying through AWS Marketplace. The 12-month contract is priced by volume of data monitored, in units of 10, with $100 overage per unit. Payment is upfront or in installments, sales handles private offers, and the same unit applies to SaaS or a deployment in my own VPC.
• Usage-based and enterprise plans (quote): for teams buying direct or through Google Cloud Marketplace, where GCP credits and consolidated billing apply.
• Free trial: the terms list one, but its length is not published, and access starts with a login URL from Telmai Support rather than a signup form.

Option	Price	Billing unit	Where it runs
AWS Marketplace Telmai 60K	$60,000 for 12 months	Data volume in units of 10; $100 overage per unit	SaaS or my own VPC
Direct or Google Cloud Marketplace	Quote only	Usage-based tiers; GCP credits on Google Cloud	SaaS, Hybrid or VPC
Free trial	Free, length not published	Set by Telmai Support	SaaS

Telmai does not define the data-volume unit publicly, so the $60,000 contract does not convert into a per-GB, per-table or per-user price.

New tenants default to 100 scans a day and 1,000 a month (UTC); scans over a limit are refused, and scheduled scans fail until the window resets. v26.5.0 lets admins set tenant scan limits. In Hybrid mode, my own Dataproc, EMR, Databricks or Fabric Spark engine runs the scans, so that compute lands on my cloud bill.

Is Telmai Good Value for Money?

• Soda: Free costs $0 with unlimited users, and Team is $750 a month, the clearest published price in this category.
• Great Expectations and Metaplane: GX Core is open source, GX Cloud has a free Developer option, and Metaplane has a free tier.
• Acceldata: lists a 30-day free trial, a clearer evaluation window than Telmai’s unpublished one.
• Monte Carlo: sells credits per monitor, quote only; the Start tier covers up to 10 users and 1,000 monitors.
• Sifflet: prices by assets monitored (up to 500, up to 1,000, 1,000+), quote only.
• Anomalo, Bigeye and Validio: publish no prices either.

Telmai’s volume-based contract pays off when the data under watch is large semi-structured or lakehouse data that per-table or per-monitor pricing would make expensive. A team watching a few dozen warehouse tables is better served by Soda’s Team plan or a free tier from Metaplane or GX Cloud.

Reviewer’s Notes: Ask for the trial with your two or three largest lakehouse or object-storage assets connected, get the definition of the data-volume unit in writing, and compare the quote against the $60,000 marketplace contract before signing.

My Experience With Telmai

There was no signup form. On the SaaS deployment, I got my tenant login URL from Telmai Support and signed in with email and password (a VPC login comes from your own administrator instead). Telmai’s homepage has Join Waitlist as its main button next to Take a product Tour, an interactive product tour, with Book a Demo and Log in in the site navigation.

Getting Access and Connecting a Source

SSO through Okta, Microsoft Entra or Auth0 is available alongside passwords, and admins add users from the Users page under Administration, with permissions set per project.

To connect a source, I opened Connections and clicked Add connection. Then I:

1. Picked the platform.
2. Named the connection.
3. Set its scope: Global, or a single project.
4. Entered the connection properties and credentials.
5. Clicked Test connection, then Create.

Create stays disabled until the connection test succeeds.

The picker covers AWS S3, Azure Blob, Google Cloud Storage, OneLake, Snowflake, BigQuery, Databricks, Redshift, Trino, SQL Server, Fabric SQL, Hive Metastore and Iceberg REST. Redshift is provisioned clusters only. Ask AI can add a connection from chat when AI is turned on, but I used the form, which shows the scope and credential fields up front.

Metaplane and Soda let a team start free without talking to sales; Telmai needs a request to Support before the first login.

Adding Assets and Running the First Scan

From the warehouse connection I added tables in batches of up to 5 tables or views per step; a custom SQL query also works as an asset. For the object-storage connection, I picked files and set the file type.

On each asset I set:

• ID attribute: always monitored.
• Timestamp attribute: needed for Redshift and Firebolt freshness and for Historical scan.
• Schedule: how often the scan runs.
• Monitored attributes: only these get profiled.

I clicked Scan now for a one-off scan, then ran Historical scan, which replays the last 10 intervals to build a baseline faster and needs both a timestamp attribute and a schedule. Results appeared on the Profiling tab (Stats, Alerts and Drilldown values), on the asset Overview and on Home, which carries the Data Quality Score.

Every scan counts toward the default cap of 100 a day, so onboarding a large batch of assets needs a higher tenant limit from an environment owner.

Building a Monitor and Routing Alerts

The monitor builder, redesigned in v26.5.0, runs in four steps:

1. Monitor logic: type, metric and alerting mode.
2. Who gets notified: the alert channel.
3. Create a ticket: Jira or ServiceNow.
4. Details: the final step before saving.

I chose Learn what’s normal, which sets thresholds from recent scans with a history of 1 to 20 scans or Automatic. Stay within a % of normal and Stay within fixed limits are the alternatives for rules that should not adapt.

Under Administration, I set up a Slack channel and a webhook channel with Digest delivery; Each incident and Scan summary are the other modes, and Microsoft Teams and email are also supported. I added a Jira ticket rule so incidents open tickets automatically. Bigeye’s connectors include PagerDuty; with Telmai, paging runs through the webhook.

Reviewer’s Notes: Run Historical scan on day one so Learn what’s normal has a baseline to work from, and keep the alert channel on Digest until the thresholds settle.

Monitors, ML Alerting and the Data Quality Score

Telmai switches on 11 out-of-box monitors as soon as a source is connected:

• Record count drifts.
• Record count correlation drifts (needs lineage).
• Freshness drifts.
• Completeness drifts.
• DQ rules violation.
• Record ID uniqueness drifts.
• Uniqueness drifts.
• Uniqueness correlation drifts (needs lineage).
• Schema drifts.
• Data drifts.
• Numeric value drift.

PII exposure monitors are off by default and must be switched on.

Custom Monitors and Row Checks

The builder offers four monitor types:

• Data quality metric: alerts on a built-in health metric.
• Custom metric: an expression up to 250 characters or SQL up to 2,000 characters.
• Row quality check: a record-level rule.
• SQL row check: a record-level rule written in SQL.

Each scan tracks total record count, record count, completeness, correctness, freshness, uniqueness and accuracy. Freshness comes from table metadata for Snowflake, BigQuery, Databricks/Delta and Hive/Iceberg; Redshift and Firebolt need a timestamp attribute, and file-based storage and generic JDBC sources get no freshness.

dbt’s four built-in generic tests (unique, not_null, accepted_values, relationships) learn no baselines, which is the gap Telmai’s learned thresholds fill. Fixed rules still belong on requirements that must hold even when history is consistently wrong.

Data Quality Score and Monitor Recommendations

The Data Quality Score is a 0-100% weighted average of four dimensions, shown in the UI since v26.2.0 (May 2026):

• Completeness: how fully values are populated.
• Validity: the lowest pass rate among the asset’s row checks.
• Freshness: 100, or 0 or a configured penalty while a freshness incident is open.
• Integrity: open incidents against a default maximum of 20.

Monitor Recommendations (v26.3.0, July 2026) suggest rules in six categories: uniqueness, range, validity, business rule, completeness and freshness. With a catalog connected, they rank by tier and critical-data-element score; AI must be on, and each draft is reviewed before saving.

I treat recommendations as drafts, not finished coverage, and the single score as a triage signal rather than proof that an asset is correct.

Lakehouse and Object Storage Coverage

Telmai reads CSV, JSON, Parquet, Avro, Delta Lake and XML files in S3, Google Cloud Storage and Azure, and detects the file type by extension. PDF works on GCP only, with Gemini enabled. Semi-structured profiling handles hundreds of nested attributes and unlimited array elements.

Iceberg, Delta and Unity Catalog

Iceberg connects through Hive Metastore on GCP and AWS deployments, or through Iceberg REST (BigLake Metastore, GCP only, added in v26.1.2 in April 2026).

Delta works through Databricks or as a storage format. Databricks reads through a cluster JDBC endpoint, with a Read from Unity Catalog option when scan jobs run in Databricks.

The Unity Catalog integration went GA in October 2026. One connection brings tables, usage, ownership and lineage, and Telmai ranks tables by downstream impact, recommends checks from each table’s baseline, and writes quality scores and incidents back to the catalog. Catalog sync also covers Atlan (v26.5.0), Alation, Actian (with score write-back) and the OneLake catalog (v26.4.0).

Telmai does not list Hudi among its formats. Acceldata names Iceberg, Delta, Hudi and Parquet, so a Hudi shop should shortlist it first.

What Is Missing

• Streaming: no Kafka or streaming source connector.
• Redshift Serverless: the connector takes provisioned clusters only.
• Connector count: the integrations page lists 250+ connectors, while the in-product picker shows about 15 platforms. Athena, MySQL, Oracle, Salesforce, SAP HANA and Teradata have separate setup guides, and availability depends on the customer agreement.
• Binary columns: BLOB, BINARY, VARBINARY and similar types are skipped.
• Hybrid exceptions: OneLake and Trino do not run on Hybrid.

Telmai covers raw files and open table formats in one tool, but file profiling is not file freshness, so late-landing files in S3 or GCS still need a separate check.

Deployment Models and Security

Telmai runs three ways, and the choice decides whether raw data leaves my environment. SaaS uses Telmai’s compute to read my sources over the network. Hybrid (v26.3.4, September 2026) keeps the control plane at Telmai while one agent host submits scans to my own Spark engine, and only aggregated metrics, alert signals and job status return. VPC puts the full stack in my AWS, GCP or Azure account, deployed with Helm and Terraform.

Model	Where Telmai runs	Does raw data leave my environment?	What I operate
SaaS	Telmai’s cloud	Read by Telmai’s compute; only derived results stored	Source access and permissions
Hybrid	Control plane at Telmai; scans on my Spark	No; aggregates and alert signals return	Agent host plus Dataproc, EMR, Databricks or Fabric Spark
VPC	My AWS, GCP or Azure account	No	The full stack, via Helm and Terraform

The Hybrid agent runs on a VM in GCP, Azure or AWS, or on Kubernetes through Helm on GKE only. It connects outbound over TCP 443, and its sizing scales with concurrent scans, not data size.

Hybrid is the middle path for teams that want customer-side compute without hosting the application; Anomalo also runs fully in a customer VPC.

What Telmai Stores and Sends to AI

Telmai stores derived results only: aggregate metrics, schema metadata, monitor and incident records, and encrypted credentials. Detection models train per customer and are not pooled.

LLM prompts carry schema metadata, operational metadata and aggregates, never record-level values, personal data or credentials. No foundation model trains on customer data, and AI can be switched off while profiling and monitoring keep running.

Compliance

• SOC 2: Telmai states compliance against the security, availability, confidentiality and privacy criteria, with the report under NDA; the report type is not published.
• Encryption: TLS 1.2+ in transit and AES-256 at rest.
• Access: SSO with Okta, Microsoft Entra or Auth0, plus MFA and role-based access.
• API keys: user-scoped since v26.3.4.

Validio lists ISO 27001 and SOC 2 Type II outright, so a procurement team that requires a named report type should request Telmai’s report early.

Remediation, Ask AI and the MCP Server

Data binning sorts every scanned record into valid and invalid bins based on row checks, and writes them to S3, Google Cloud Storage, Azure or a Databricks Delta schema, as JSON, CSV or Parquet. It needs ID attributes and at least one enabled row check, and it runs on the next scan. I set it up from the Remediation & actions card on the asset Overview.

The circuit breaker is an API pattern, not a packaged operator:

1. Trigger a scan.
2. Wait for job completion by polling or webhook.
3. Check for alerts.
4. Stop or continue the pipeline.

I wire that into Airflow or another orchestrator myself, since there is no packaged Airflow operator or dbt package. For a hard stop on a handful of rules, dbt’s four generic tests or open-source GX Core cover it with no contract.

Ask AI, Incidents and MCP

Ask AI (v26.5.0) is an in-product agent that:

• Adds connections and assets.
• Summarizes recent changes.
• Describes an asset.
• Runs Investigate with AI on an incident, returning the most likely root cause and a next step.

Telmai’s homepage presents Ask AI’s capabilities as seven Data Reliability Agents: Orchestration, Validation, Incident Diagnosis, Lineage, Data Insight, Help and Routing.

Incidents also show blast radius from lineage. Environment owners must turn Ask AI on, requests consume tokens against tenant limits, and no chat history is kept.

The Telmai MCP server lets Claude and clients such as Cline query Telmai per tenant, with OAuth through the tenant’s SSO.

Binning is the feature that sets Telmai apart from alert-only tools; Ask AI’s root cause is a lead, not a diagnosis.

How Does Telmai Compare to Competitors?

• Monte Carlo: credit-based pricing per monitor, quote only, across Start, Scale, Enterprise and Business Critical tiers; Start covers up to 10 users and 1,000 monitors, with no free tier or trial. It wins for warehouse-centric teams that budget per monitor rather than by data volume.
• Bigeye: no public price; lineage connectors include Airflow, dbt, Snowflake, Databricks, BigQuery, Tableau, Power BI, Slack, Jira and PagerDuty, plus monitoring as code through bigconfig. It wins for lineage-led enterprise programs that page through PagerDuty.
• Anomalo: SaaS or fully in my VPC, unstructured data monitoring and SOC 2 Type II, with no public price. It wins when unstructured data needs quality checks too.
• Sifflet: tiers by assets monitored (Entry up to 500, Growth up to 1,000, Enterprise 1,000+), quote only, and it can use Snowflake credits. It wins for Snowflake buyers.
• Metaplane: a free tier, Pro priced per monitored table and custom Enterprise. It wins for small teams that want to start free.
• Acceldata: a 30-day free trial, cloud, on-prem, hybrid or Kubernetes deployment, and Iceberg, Delta, Hudi and Parquet support. It wins for on-prem or Hudi estates.
• Soda, Great Expectations and dbt tests: Soda Free at $0 and Team at $750 a month, GX Core open source with a free GX Cloud Developer option, and dbt’s four generic tests. They win for teams that want checks as code with no contract.
• Validio: a 14-day free trial for up to 10 users, ISO 27001 and SOC 2 Type II, no public price. It wins when a fixed evaluation window and a named report type both matter.

Telmai wins where most of the data under watch sits in Iceberg, Delta or raw files in object storage and must stay in my cloud.

How I Reviewed Telmai

I worked in Telmai v26.5.0 on the SaaS deployment. I added a warehouse connection and an object-storage connection, added assets, ran a scan and a Historical scan, built a monitor with Learn what’s normal, set up Slack and webhook alerts and a Jira ticket rule, configured data binning, and tried Ask AI and Monitor Recommendations. I also went through the Hybrid and VPC setup requirements and the security controls, and priced Telmai against the ten rivals compared above.

I weighted:

• Pricing and value: contract terms, trial access and scan limits.
• Setup and first scan: connections, assets and Historical scan.
• Monitors and alerting: learned thresholds, channels and tickets.
• Lakehouse coverage: Iceberg, Delta and raw files.
• Deployment and security: SaaS, Hybrid and VPC.
• Remediation and AI: binning, circuit breaker and Ask AI.
• Alternatives: free entry versus quote-only contracts.

Prices current as of November 2026.

Telmai Review: Should You Use Telmai for Data Observability?

Yes, if much of your data sits in Iceberg, Delta, Parquet or semi-structured files in S3, GCS or Azure alongside Snowflake, BigQuery or Databricks, and policy keeps row data in your cloud. Hybrid and VPC handle that boundary, and Spark profiling, learned thresholds and data binning do work that warehouse-only tools leave to you.

No, if you watch a few dozen warehouse tables and want to start free; Soda, Metaplane and GX Cloud get you there without a sales call. Skip Telmai too if you need Kafka or streaming monitoring, or if per-monitor pricing and native PagerDuty are the priority, where Monte Carlo and Bigeye fit better.

Settle three things before signing:

• Unit: what one data-volume unit means and how the $100 overage applies.
• Trial: its length and which assets it covers.
• Hybrid compute: the Spark cost that lands on your own cloud bill.

Decide up front whether the missing freshness monitor on S3, GCS and Azure files is acceptable, because no Telmai monitor computes it for files. Bring your largest object-storage and lakehouse assets to the first call, and request access at Telmai.

FAQ

How much does Telmai cost?

Telmai has no public price list; it sells usage-based plans by data volume, quote only. The one published figure is the AWS Marketplace contract: $60,000 for 12 months, priced in units of 10, with $100 overage per unit and an undefined unit. Google Cloud Marketplace is another route, with GCP credits and consolidated billing. On Hybrid, the Spark compute also lands on your own cloud bill.

Does Telmai have a free trial?

Yes, Telmai’s terms list a free trial, but its length is not published. There is no self-serve signup: SaaS access starts with a tenant login URL from Telmai Support, and the AWS Marketplace offer shows no trial. Acceldata (30 days) and Validio (14 days) publish fixed windows.

Does Telmai keep my data in my cloud?

Hybrid and VPC keep row-level data in your environment. Hybrid sends only aggregated metrics, alert signals and job status to Telmai’s control plane, and VPC runs the full stack in your AWS, GCP or Azure account. SaaS reads sources with Telmai’s compute, and every model stores only derived metrics and metadata.

Does Telmai support Iceberg and Delta Lake?

Yes. Iceberg connects through Hive Metastore on GCP and AWS deployments, or Iceberg REST on GCP only. Delta works through Databricks or as a storage format, and Telmai also reads Parquet, Avro, CSV, JSON and XML files. Telmai does not list Hudi, so Acceldata is the better fit for Hudi tables.

What are the best Telmai alternatives?

Monte Carlo and Bigeye suit warehouse-centric and lineage-led programs, Anomalo covers unstructured data, Sifflet suits Snowflake buyers, Metaplane starts free, and Acceldata covers Hudi and on-prem. For checks as code with no contract, Soda, Great Expectations and dbt tests are the free route. See How Does Telmai Compare to Competitors? above.