Build, version, and ship complex automation visually. Every workflow compiles to a typed, schema-checked graph that runs across your worker fleet — with first-class control flow, edge transforms, and strict data contracts.
Drag-and-drop SVG canvas with a node toolbox and a flyout config panel. Layout lives inside the workflow definition — no sidecar files, no drift between editor and runtime.
Every edge between nodes can carry a CEL expression that transforms, filters, or reshapes data in transit. Type-checked against field schemas at compile time, with a built-in snippet library.
Sequential, Parallel, Loop, Condition, Switch, Batch, and Split — real control flow primitives with worker pinning, iteration safety limits, CEL exit checks, and configurable output modes.
JSON-schema-based field definitions with typed fields — string, number, boolean, file, url, email, date, json. Required flags, list fields, and schemas that are reusable across every trigger and agent node.
First-class source and sink nodes for BigQuery, PostgreSQL, and MySQL. Credentials live in an encrypted vault and integrations declare whether they provide reads, writes, or both — so the editor only shows you legal wiring.
// Runs on the edge between Validate → Enrich.
// Compiled once, executed per message.
{
"user_id": input.customer.id,
"amount": input.order.total * 1.08,
"region": input.customer.address.country,
"tags": input.order.items.map(i, i.sku),
"priority": input.order.total > 1000 ? "high" : "normal"
}A Nanomite worker is a single binary that turns any machine into an execution node on the grid — from the laptop in your backpack to the rack in your datacenter. Nothing phones home except the worker itself.
Run workflows locally during development with the same binary you ship to production. No special mode, no emulator — your laptop is just another worker on the grid.
Drop workers on the servers that already sit next to your data. No network egress, no cloud data tax, no surprise bills from a vendor moving bytes across regions.
Ship workers to edge devices near the sensors, cameras, and machines that generate your events. Process where the data lives — send only decisions back to the control plane.
Spin up GPU-backed workers in your own cloud accounts. Nanomite never touches your VPC — it just connects to whoever dials in. Your account, your IAM, your bill.
Agents are first-class citizens in Nanomite — multi-provider, multi-modal, and composable. Package expertise as skills, plug in external tools via MCP servers, and wire them into workflows with typed inputs and outputs.
Multi-provider out of the box — Anthropic, OpenAI, Google, xAI, and local LLMs via Ollama and vLLM. Chat or tool-style. Swap providers mid-conversation without losing context.
Package domain knowledge as reusable, markdown-based skills. Inject expertise into any agent without copy-pasting prompts. Version skills like code.
Plug in any Model Context Protocol server — streamable HTTP, SSE, or local stdio. Auth headers stay in the vault. Cached tool discovery with JSON Schema validation.
Nanomite ships with a growing catalog of integrations — unified source and sink semantics, credentials in an encrypted vault, and editor affordances that only let you wire what actually works.
Service Account or Workload Identity auth. Storage Write API sink, managed writer.
Host, port, SSL mode, schema lock. Source tooling for agents, batch INSERT sink.
TLS mode, database lock, schema discovery. Source tooling for agents, batch INSERT sink.
Low-latency video stream ingestion straight into workflows. Pair with on-prem ML models for real-time video intelligence: camera tracking, signal analysis, object detection, and live event decisioning — all processed on workers next to the cameras.
Every LLM call in your workflow is already labeling data — prompts in, judgments out. We'll let you turn that signal into your own fine-tuned models, trained on your examples and served from your own workers via Ollama or vLLM. Every run feeds the next retraining cycle, so the system gets cheaper, faster, and more private the more you use it. Ideal for complex routing, data anonymization, ticket triage, field extraction, and any narrow task where a trillion-parameter general model is overkill.
And you don't have to wait for workflow history to build up. Point a training job at any connected data source — BigQuery, Postgres, MySQL, or any other integration you wire up — and workers pull labeled examples straight from your systems of record. Historic ground truth in your warehouse becomes a cold-start for a private model, and none of it ever leaves your perimeter. That's how you actually ship ML in health, finance, legal, and gov.
Stream ingestion at AWS scale. Bind shards to workers, process records in flight.
Managed messaging with push and pull semantics. At-least-once delivery into workflows.
The event streaming backbone. Consumer groups, exactly-once, compacted topics.
SMS, voice, and WhatsApp as both triggers and actions. Conversational automation at scale.
Classic AMQP messaging with queues, exchanges, and routing keys — on your own broker.
Snowflake, Databricks, Redis Streams, NATS, MQTT, S3 event notifications. Tell us what you need.
Need something that isn't on the list?
Request an IntegrationWorkflows, workers, agents, and integrations are one platform — all designed to run on infrastructure you own.