Subscribe to MQTT, NATS, and AMQP streams — sensors, vehicles, machines — and turn each event into a governed, multi-step AI workflow. No bridge service, no orchestration code.
Model flows as a graph — triggers, agents, tools, and human approvals on one canvas. Captured from Arcflux Studio: a real alert-triage flow, built, published, and run. No mockups.
01Build
Sketch the flow on the canvas
Webhook in, dedupe gate, Claude triage, Linear out — five nodes, no orchestration code. Trigger it from webhooks, schedules, chat, the API, or live device streams.
02Run
Watch every step stream
Tokens, timings, and the drafted Linear issue — live while the run executes. Switch the model per step without rewriting the flow.
03Govern
Humans approve the risky steps
Runs pause at approval gates, and policy checks drop noise before a run ever starts. Nothing sends until someone clicks Approve.
studio — databricks data pipeline · published v2
test run — 4 steps · completed
approvals — 3 pending
// How it works
Real workflows, really running
Model flows as a graph — triggers, agents, tools, and human approvals on one canvas. Captured from Arcflux Studio: a real alert-triage flow, built, published, and run. No mockups.
01Build
Sketch the flow on the canvas
studio — databricks data pipeline · published v2
Webhook in, dedupe gate, Claude triage, Linear out — five nodes, no orchestration code. Trigger it from webhooks, schedules, chat, the API, or live device streams.
02Run
Watch every step stream
test run — 4 steps · completed
Tokens, timings, and the drafted Linear issue — live while the run executes. Switch the model per step without rewriting the flow.
03Govern
Humans approve the risky steps
approvals — 3 pending
Runs pause at approval gates, and policy checks drop noise before a run ever starts. Nothing sends until someone clicks Approve.
// Use cases
Workflows you could ship this week
Each one starts from a real event — a device stream, a webhook, a message — runs agents in the middle, and ends in a governed action.
01
mqtt → work order
Fleet & telematics
A vehicle's MQTT telemetry crosses a threshold. The workflow correlates recent readings, asks an agent to check the maintenance runbook, and drafts a work order — paused for human approval before dispatch.
02
mqtt → mes
Factory floor (OT↔IT)
Machine signals over MQTT reach an agent that diagnoses the fault, logs it to your MES, and escalates only real anomalies. Policy drops sensor noise before it spends a credit.
03
nats → incident
Robotic fleets
Robot fault events stream over NATS. An agent triages, drafts the incident with full context, and routes high-risk stops to a human before anything actuates.
04
webhook → linear
Alert triage
A monitoring webhook fires. The workflow dedupes against open incidents, asks an agent to check the runbook, and files a Linear issue with full context. Duplicate alerts are dropped before they spend a credit.
// Who it's for
Built for teams that run on real events
Whatever fires — a sensor, an alert, a message — Arcflux turns it into a governed, multi-step AI workflow. Three teams get there first.
Ops & platform teams
Run the physical world
Fleet, factory, robotics. Subscribe to broker topics the way you'd add a webhook — wildcards, QoS, connector health. Sensors and machines trigger workflows directly, no bridge service to babysit.
mqttsensors/7/temp→agent→work order
MQTTNATSAMQP
DevOps & SRE
Tame the alert firehose
Monitoring fires, policy gates dedupe and throttle, an agent checks the runbook and files the issue with full context. Duplicate alerts drop before they spend a credit.
webhookPOST /alerts→triage→linear
WebhookCronAPI
Product & ops builders
Automate the busywork
Customer support, lead qualification, document processing. Agents draft the work; risky steps pause at an approval gate. Nothing sends until someone clicks Approve.
chat#support→draft→approval
ChatScheduleWidget
Every source, one policy gate
HTTPMQTTNATSAMQPCRONCHATCoAP · soonMAVLink · soon
Already run a broker? Keep it. Arcflux subscribes to MQTT, NATS, and AMQP natively — it doesn't replace your broker, it turns its events into governed AI action.
// Integrations
Your tools, already wired in
Bring any MCP server or native integration. Tools are scoped per workflow.
Govern every event before it runs
// Govern
Channel routing, filters, policy checks, and human approval sit between a source and a run.
5 checks
dedupe, cooldown, throttle, condition, safety
100%
of runs auditable, step by step
<100ms
SLA pickup, trigger to run
// Pricing
Pricing that scales with your runs
Basic is free with 100 credits a month. Upgrade when your workflows leave the workshop.
Answers to common questions about Arcflux and agentic workflows. Anything else: info@arcflux.ai.
A workflow is a graph of steps (LLM calls, tools, knowledge lookups, and human approvals) that runs automatically when a real-time event fires. Arcflux lets you build one visually, with no orchestration code required.
Every event passes a policy layer before it runs; skipped events spend zero credits, so noisy sources never burn your quota. A credit is roughly 1 cent of LLM usage: a simple run costs about 1 credit, a multi-step run with knowledge lookups 2 to 5.
From webhooks, schedules, chat, the API or CLI, an embeddable widget, or device streams over MQTT, NATS, and AMQP. One workflow can listen to many sources at once.
Claude, GPT, Gemini, Grok, and Groq today. You can switch the model per step without rewriting the workflow.
Yes. Bring any MCP server or native integration such as Linear, Slack, Stripe, and databases. Tools are scoped per workflow.
Workflows run with per-flow scoping and run-level auditing. SSO/SAML and dedicated data isolation are on the enterprise roadmap.
Turn your next event into a workflow
Model agentic workflows visually, trigger them from anywhere, and govern every run.
