TDengine vs IronFlock: Industrial Data & AI Platform Comparison (2026)

TDengine is one of the few industrial platforms that, like IronFlock, treats AI as part of the product rather than as a bolt-on. It started as a high-performance time-series database — open-sourced in 2019, written in C, AGPL-3.0, with more than a million installations and 25,000+ GitHub stars — and has since grown upward into a full data stack: TDengine TSDB for storage, TDengine IDMP (Industrial Data Management Platform) for asset modeling, dashboards and events, and an AI layer made up of TDgpt (forecasting and anomaly detection invoked directly from SQL) and the Industrial Agent Runtime (prebuilt AI assistants plus an MCP interface that exposes platform capabilities to external agents). In 2026 the company repositioned from “high-performance time-series database” to “industrial data foundation for the AI era.”

IronFlock approaches industrial AI from the opposite end. Rather than starting at the database and reaching toward the assets, IronFlock starts at the device — a lightweight agent and Docker-containerized apps running autonomously on any Linux or Windows hardware — and reaches upward into central services: FleetDB (TimescaleDB) for fleet-wide time-series data, dashboarding, alarms, and an AI service whose agents can act on physical devices, not only read from them.

That difference matters more than any single feature. TDengine is an outstanding answer to “where does my industrial data live, and how do I ask questions of it?” It is not an answer to “how do I get software onto 800 machines in the field, update them, reach them remotely, and let them keep working when the network drops.” IronFlock is built for the second question and provides the data and AI layer on top of it.

At a Glance

Key Differences

Architecture

TDengine: A Three-Layer Data Stack

TDengine describes its own architecture as three layers, and that is a fair description of how it is built:

• Layer 1 — TDengine TSDB: The time-series database. Written in C, one table per data collection point grouped into supertables, with RAFT-based clustering, separated storage and compute, Kubernetes deployment, and tiered storage in the Enterprise edition. It also ships caching, stream processing and data subscription, so a lot of what teams normally bolt on around a TSDB is inside the database. taosAdapter exposes REST, InfluxDB line protocol, OpenTSDB, Prometheus and Telegraf compatibility.
• Layer 2 — TDengine IDMP: The industrial data management platform. IDMP stores no time-series data of its own — it sits above TSDB (or another database) and adds a tree-structured element hierarchy of factories, lines, machines and measurements, with attributes, relationships, analytics, dashboards and events attached to each node. This is the semantic layer that turns raw tags into a digital twin.
• Layer 3 — Open interfaces for AI: TDgpt runs analysis nodes (anodes) that serve forecasting, anomaly detection and imputation to SQL queries, backed by statistical algorithms, machine learning and time-series foundation models (TDtsfm, Time-MoE). The Industrial Agent Runtime adds thirteen prebuilt industrial assistants, Markdown-defined “skills,” a semantic knowledge base, and an MCP server that exposes 50+ platform capabilities as tools to external AI agents.

Data gets in through taosX, the Enterprise zero-code ingestion pipeline, and taosX-Agent, which runs at the edge and speaks OPC UA, OPC DA, MQTT, Kafka, PI System, AVEVA Historian, CSV/Parquet and legacy TSDB formats. The agent can filter and preprocess before transmission and buffers locally with store-and-forward when the link to the central system drops.

The important thing to understand about the edge in this architecture: taosX-Agent is a data collector, not a compute platform. It adapts protocols, filters, and forwards. Your control logic, your vision model, your local HMI and your custom application have to live somewhere else.

IronFlock: Distributed Edge + Central Services

IronFlock is a distributed system with two complementary layers. Autonomous edge devices run a lightweight agent and Docker-containerized apps at the point of operation. Central services — FleetDB (TimescaleDB), the FleetDB Service, the AI Service, and the web UI — provide fleet-wide storage, dashboarding, alarms and intelligence. A WAMP message broker connects everything with real-time pub/sub and RPC, enforcing cryptographic isolation between projects.

• Edge devices: Any Linux- or Windows-capable hardware — Raspberry Pi, industrial PCs, NVIDIA Jetson, Windows IPCs, gateways. On Windows the agent runs as a native service with automatic restart and self-update.
• Apps: Docker containers in any programming language, deployed to edge devices or to virtual devices — cloud-hosted compute nodes that join your project alongside physical hardware and run fleet-wide services like Grafana, Node-RED or Jupyter.
• Data: Edge apps publish telemetry through the broker to FleetDB, which auto-provisions per-project TimescaleDB tables you can query with plain SQL.
• AI: The AI Service orchestrates multi-agent conversations that read fleet data, generate charts and dashboards, and invoke functions on physical devices. Apps can ship their own agent definitions as YAML templates.
• Deployment: Cloud SaaS or on-premises — the full platform runs in your own infrastructure.

See Architecture for the full breakdown.

What This Means in Practice

Feature Comparison

Data & Connectivity

AI & Analytics

Both platforms ship AI over industrial data, and both answer questions in natural language. The difference is what happens after the answer: TDengine reports it, IronFlock can also act on it.

Visualization & Dashboards

Device & Fleet Management

This is where the two products stop overlapping: device management sits outside TDengine’s scope, so everything in this table has to come from somewhere else in a TDengine deployment.

Remote Access & Security

App Development

Alarms & Notifications

Pricing Comparison

TDengine: Free Up to 5,000 Tags, Then Licensing per Tag

Since July 2026, TDengine’s free tier covers the complete platform — TSDB plus IDMP, all 20 data connectors, asset modeling, dashboards, analytics and the AI features — for up to 5,000 tags, in production, with no ingest or query throttling and no trial deadline. A tag is one continuously monitored signal (asset or device plus measurement name), independent of sampling rate; TDengine sizes 5,000 tags at roughly 100–200 machines, or one moderately complex production line. The license renews free, indefinitely, though activation and renewal need temporary internet connectivity. Two caveats are worth knowing: support is a paid option rather than part of the tier, and third-party query access to the bundled TSDB requires a separate commercial license.

Above 5,000 tags, licensing is per tag tier, self-hosted, as either an annual or a perpetual license, with surcharges for high availability and, on perpetual licenses, for ongoing support. TDengine Cloud is priced separately as a managed service. The AGPL-3.0 TSDB-OSS remains genuinely free and self-hostable, but without IDMP, without taosX connectors, without encryption, and without user privilege controls.

The model is clean and predictable, and notably does not punish high sampling rates — a strong fit for deep history on a fixed set of signals. It does scale with the breadth of your instrumentation: every new measurement on every new machine is another tag, whether or not anyone ever looks at it.

IronFlock: Free Cloud + Subscription for On-Premises

IronFlock’s cloud version is free — device management, dashboards, data storage, OTA updates, alarms, remote access and app deployment are included at no cost. Charges follow resource usage: storage, remote access sessions, virtual devices and AI usage. There are no tag counts, no point licenses, no per-device fees and no mandatory support contract. See the pricing page for details.

Additional capabilities can be added by purchasing apps from the marketplace — protocol connectors, analytics tools or industry-specific solutions built by IronFlock or third-party developers. For on-premises deployments (air-gapped or private infrastructure), IronFlock offers a subscription-based license.

When to Choose TDengine

TDengine is the better choice if:

• The database is the point. Your workload is extreme by TSDB standards — millions of tags, very high sampling rates, years of retention on one cluster — and a purpose-built engine is worth giving up a general-purpose SQL dialect for.
• You are migrating off a historian or another TSDB — the zero-code PI System, AVEVA, InfluxDB and OpenTSDB connectors make that move cheaper than writing it yourself.
• You want a bundled time-series foundation model. TDgpt runs TDtsfm or Time-MoE from inside a query, with no model of your own to choose, host or maintain.
• Your devices are already connected and already managed — an existing SCADA, PLC or gateway estate that produces data reliably, where the missing piece is a modern place to put it and a way to reason about it.
• You want a semantic asset hierarchy over a large plant: elements, attributes, relationships and a digital twin model that AI agents can navigate.
• You want a genuinely open-source core you can self-host and inspect (AGPL-3.0), with the option to stay in the free tier for a small plant indefinitely.
• You want to expose your industrial data to external AI agents over MCP and let your own agent stack do the reasoning.

When to Choose IronFlock as a TDengine Alternative

IronFlock is the stronger choice when:

• You have a fleet, not just a database. Provisioning, grouping, OTA updates across OS, agent and applications, live logs and device health are the daily work of running machines in the field — and none of it is in TDengine’s scope.
• You need real edge computing — applications in any language running on the machine, doing control, vision, local buffering and local HMI, not just protocol adaptation and forwarding.
• You need remote access without VPN — HTTP, SSH, VNC, TCP and UDP tunneling built into the platform, for every device, from the browser.
• You want AI that acts, not only advises. IronFlock’s agents invoke functions that device apps expose, under the calling user’s privileges — which means an AI conversation can change a setpoint, restart a process or trigger a routine, not just tell you that something looks wrong.
• Your sites must keep working when the network is down — IronFlock devices continue running every app autonomously, serving local screens and collecting data, and reconcile when they reconnect.
• You are an OEM shipping digital services to your own customers and need per-customer data isolation and a white-label portal, not a single-tenant plant database.
• You want to build and monetize apps — package your domain expertise as a container and sell it through an open marketplace.
• You want SCADA-grade visualization — a full P&ID symbol library, action widgets and forms — alongside the analytics.
• Tag counting doesn’t fit your data. Broad instrumentation across many machines — hundreds of signals each, most of them looked at only when something goes wrong — gets expensive quickly when every signal is a licensed tag.
• You need to answer business questions, not just tag questions. FleetDB is full PostgreSQL: join production counts against quality results and energy readings from three different apps, express OEE or yield as one query, and save it as a transform that keeps itself current and feeds boards and the AI like any other table.
• You want one system instead of a stack. Storage, dashboards, SCADA graphics, alarms, device management, remote access and AI come from the same platform, with one access model over all of it — rather than a database layer that still needs a device layer underneath it.

Migration Path

IronFlock can run alongside an existing TDengine deployment while you move across. Start at the machine: an IronFlock collector app reads the same PLCs over S7, Modbus or OPC UA on the edge device and writes into FleetDB, so new dashboards, alarms and AI conversations are built on IronFlock while the old TSDB keeps serving whatever still points at it. Historical data comes over in bulk through a containerized migration app against TDengine’s REST interface.

From there the device layer is what changes the picture — provisioning, OTA updates, remote access and offline operation arrive for the same machines that were previously only sending tags — and the tag license stops growing with every measurement you add.

Ready to try it? Start free — connect a device and see your first dashboard in minutes.