icp/ai-data

Feed live device data to your AI agents and data platform.

UltraMQ is an MQTT broker for AI and data teams. It streams device data into Kafka, your lakehouse and your agents without glue code, and lets agents send commands back to machines when they have permission.

01ultramq/ai-data/problems

What makes device data hard to use. Four problems we hear from data and AI teams.

01

Glue code between MQTT and Kafka

Every team writes its own bridge service, with its own retry logic, topic mapping and monitoring.

02

Schemas drift

Firmware updates change payloads without notice. Pipelines break downstream, often at night.

03

Agents only see yesterday

Most agent tooling reads from batch tables. For maintenance or control decisions, the context needs to be live.

04

Latency and cost per message

Some services bill per message or per connection-minute. At device scale, those costs are hard to predict.

02ultramq/ai-data/capabilities

What UltraMQ does for your pipeline. From device to model and back.

01

Streaming into your platform

Native sinks move topics into Kafka, lakehouses and analytics tools without a custom bridge. [Integrations list to confirm]

02

Live topics agents can read

Agents subscribe to live topics for real-time context instead of polling tables. [Agent interface (MCP and/or API) to confirm]

03

Commands back to machines

With the right permissions, agents publish commands to machine topics, closing the loop from insight to action. [Permission model to confirm]

04

Rules in flight

Filter, enrich and route messages before they reach your platform, so you only store what you need. [Rule engine to confirm]

05

Reproducible benchmarks

We will publish the method and setup so you can run the numbers yourself. [Benchmarks pending]

06

Fixed price per connection

You pay per connected device, so cost follows your fleet size. [Pricing model and message limits to confirm]

Values in [brackets] are not confirmed yet.

03ultramq/ai-data/architecture

How the loop works. From a sensor on line 3 to an agent and back.

  1. 01

    Publish

    Devices publish telemetry over MQTT to UltraMQ.

    PUBLISH plant/line-3/temp
  2. 02

    Stream

    Topics stream into Kafka and your lakehouse for history and training.

    SINK kafka://plant-events
  3. 03

    Reason

    An agent subscribes to live topics and combines them with history to decide.

    SUBSCRIBE plant/+/temp
  4. 04

    Act

    With permission, the agent publishes a command that the machine executes.

    COMMAND cmd/line-3/slow-down
04ultramq/ai-data/team

What each person on your team gets. We go through it with you in the first call.

RoleTheir questionWhat we work through with you
AI or data engineerHow fast can I get data flowing?Your data sources, topic mapping, a first pipeline into your stack
Platform / SREHow do we run and observe it?Deployment option, scaling, monitoring [observability features to confirm]
Head of data / ML platformDoes it fit our architecture and budget?Integrations, pricing per connection, roadmap fit
SecurityWho can send commands to machines?Authentication, topic permissions, audit [security features to confirm]
05ultramq/ai-data/faq

Frequently asked questions

How do I get MQTT data into Kafka?

Common options are a Kafka Connect MQTT connector, a bridge service you write yourself, or a broker that streams topics to Kafka natively. UltraMQ is designed for the native route [integrations to confirm].

Can AI agents read live MQTT data?

Yes. An agent can subscribe to topics like any other MQTT client and receive messages as they are published. UltraMQ plans a dedicated agent interface [MCP and/or API to confirm].

Can an agent send commands back to a machine?

Yes, if it has permission to publish to that machine’s command topic. Who may publish where is controlled by topic-level permissions [permission model to confirm].

Why use MQTT and Kafka together?

MQTT is built for many devices on unreliable networks; Kafka is built for durable, replayable streams and analytics. A broker at the edge of the network feeds Kafka in the data centre.

Are there public benchmarks?

Not yet. We will publish the method, the setup and the results so you can reproduce them. Until then, all figures on this site are marked as illustrative.

07ultramq/start

Talk to an engineer about your project.

Tell us what you want to connect, where it has to run and what the data should feed. We reply by email.

SUBultramq/#