Agent-native backend
Your agent builds the backend it needs, then fills it.
The schema is runtime data, so templates, typed attributes, and the first records are all created through the API, with no migration and no deploy.
Point Claude, Cursor, or your own agent at Omnismith over MCP. It designs the template, picks the attribute kinds, creates the records, and streams metrics into them. What it builds carries change history, charts, role-based access, and a REST API from the first write.
Multi-site facility telemetry
Workspace iot-fleet
What you ask for
MCP sessionWe run refrigeration units across 40 retail sites. I need to track the site, the install date, the service tier, and a coolant temperature reading every five minutes.
Created the template chiller_unit in iot-fleet with six attributes. Coolant temperature and compressor hours are metrics, so each one carries its own time series on the record.
Add the unit at Rotterdam Noord and load today's readings.
Created CHL-RTM-04 and ingested 288 coolant temperature observations. Latest reading is 4.4 °C.
Tool calls
- create_template chiller_unit
- create_attribute 6 attributes, 4 kinds
- create_entity CHL-RTM-04
- ingest_entity_metrics 288 observations
What it builds
chiller_unitChiller unit
- Site Points at retail_site Reference
- Install date Date Dimension
- Service tier Standard · Extended · 24/7 List
- Coolant temperature Number · °C, every 5 minutes Metric
- Compressor hours Number · hours Metric
- Status Healthy · Drifting · Offline List
First record
CHL-RTM-04
- Site
- Rotterdam Noord
- Install date
- 2024-03-18
- Service tier
- Extended
- Coolant temperature
- 4.4 °C
- Compressor hours
- 11,420
Coolant temperature
288 observations today, held on the record itself and queryable by time range.
Launch with a Sentence
Tell Omnismith what to build.
Choose a commercial starter prompt or write your own custom sentence. Omnismith pre-configures your workspace schema the moment you sign up — no empty canvas.
Why an agent can do this here
The schema is data, so it can be written at runtime.
In a fixed-schema tool the columns are part of the build. Changing them is a migration, a review, and a release window, which puts schema design outside what an agent can do during a conversation. In Omnismith a template and its attributes are records, created and changed through the same API that creates rows. That is the whole difference, and everything on this page follows from it.
Runtime schema
A new field is one API call.
POST /v1/attributes defines an attribute and returns its id. The next write can use it. There is nothing to generate, compile, or ship in between.
Four typed kinds
The type system tells the agent what a field is for.
Dimension holds current state. Metric holds a time series. List holds a controlled option set. Reference points at another record. Eight storage types sit underneath: string, number, boolean, datetime, date, file, image, and markdown.
Governed from the first write
The agent works inside the same controls your team does.
Dimension changes land in the append-only history log with author and timestamp. Metric observations stream into time-series storage. Role-based access applies to an agent session exactly as it applies to a person.
What the agent actually sends
Adding a metric to a live template, mid-conversation.
attribute_type: 1 is a metric, so the field arrives with time-series storage already attached. Every record on the template can start accepting observations on the next call.
One job, end to end
Multi-site facility telemetry, from the first template to the alert that wakes someone up.
Chillers, pumps, HVAC, and sensor nodes spread across sites are the clearest fit: the job is equipment records plus readings over time, and the person who owns the equipment budget is the person who feels the gap. The four stages below are the whole job.
Describe the fleet
01
The agent writes the model from a description of the equipment.
One template per equipment class, a reference to the site it stands in, a controlled list for service tier and status, and a metric for every reading worth trending. The model exists as soon as the conversation ends.
Point the devices at it
02
Readings arrive as batched observations on the record itself.
POST /v1/entities/{id}/metrics accepts a batch of observations against any metric attribute. Gateways, scripts, and scheduled jobs write straight into the same record an operator opens in the browser.
See the drift
03
Trends stay attached to the unit that produced them.
Per-entity charts aggregate with avg, sum, min, max, count, first, or last over buckets from one second to one month. Dashboards assemble stats, charts, gauges, and filtered lists across the whole fleet.
Act on the threshold
04
A crossed limit turns into a message to whoever owns the site.
Automations evaluate multi-attribute conditions on entity and metric events, then dispatch to Telegram, webhooks, or mobile push. The record that triggered the alert carries the history that explains it.
What the facilities owner ends up with
One system holding the asset, its readings, and the record of who changed what.
- Every unit, its site, and its service tier in one queryable model
- Field-level change history with author and timestamp for audit
- Metric retention and charting without a separate monitoring account
- Role-based access so contractors see their sites and nothing else
What the agent hands you
Whatever it builds is production-grade from the first record.
The agent chooses the shape. Everything around that shape is already in place and applies to every template it creates, with no configuration step in between.
History
Every dimension change is kept, with the name attached.
The append-only history log records the previous value, the new value, the actor, and the timestamp for each field. Filter it by attribute, date window, author, or value.
Immutable audit historyCharts
Metric history lives on the record it describes.
Metric attributes carry their own time series. Chart any of them per entity with avg, sum, min, max, count, first, or last, over buckets from one second to one month.
Access
Roles apply the same way to people and to agent sessions.
Owner, Admin, Editor, and Viewer roles gate every write, and dynamic entity scopes narrow what a member reaches. An MCP session inherits the permissions of the person who authorized it.
Role-based access controlAPI
The REST API exposes the structure the agent just created.
Programmatic access tokens read and write the same templates, attributes, and entities. The OpenAPI specification describes all of it, which is what makes the tool catalog possible.
OpenAPI specificationAutomations
A crossed threshold becomes a message.
Automation rules evaluate multi-attribute conditions on entity lifecycle and metric events, then dispatch to Telegram, webhooks, or mobile push.
Webhook automationsSearch
Records stay findable as the schema grows.
Structured filters across dynamic attributes, full-text search over string fields, and semantic search for the cases where the exact wording is unknown.
Rules
A status change is either valid, or it never happened.
When/then rules are enforced on every write, and named Actions bundle a precondition, the fields an operator fills in, and the values set silently into one guided operation.
Rules & ActionsDeveloper surface
The API follows the same structure your team sees in the UI.
Use programmatic access tokens to read and update the same templates and entities from scripts, integrations, or a custom frontend. Attributes can be addressed by slug, so a payload written today survives an id change.
Connect it
Two minutes from a Claude session to a workspace it can build in.
Omnismith runs a Model Context Protocol server with a curated tool catalog and OAuth 2.1 authorization. Every call the agent makes executes as you, against the one project you approved, under your role permissions, and lands in the audit log with your name on it.
Create a free workspace
250 entities, 25 attributes, and 1,000 AI credits, with no card.
Add the server to your MCP client
Paste the URL into Claude Desktop's custom connector dialog, or add it to the configuration file your client reads.
Approve the browser consent screen
OAuth 2.1 opens in your browser. You pick which project the client may reach, and the session is scoped to it.
Ask for the model you need
The agent designs the template, defines the attributes, and creates the first records inside your role permissions.
Client configuration
Claude Desktop
Open Settings → Connectors → Add custom connector, then paste the server URL.
Cursor and other config-file clients
Add the server to the MCP configuration your client reads.
{
"mcpServers": {
"omnismith": {
"url": "https://mcp.omnismith.io/mcp"
}
}
} Either route opens the Omnismith consent screen in your browser. You choose the project, and the tools appear in the session.
Some of what the agent gets
- scaffold_data_model
- create_attribute
- create_entity
- ingest_entity_metrics
- search_entities
- get_entity_chart
- create_automation
The catalog covers schema design, records, telemetry ingestion, search, dashboards, and automations, with workspace schemas exposed as live resources the agent can read before it writes.
Starting points
Install a blueprint, then tell the agent how your version differs.
A blueprint installs a working set of templates, attributes, and sample records in a single call. It exists to give the agent something to reshape: the schema is still runtime data afterwards, so the first customisation is another sentence.
Featured blueprint
Smart Facility Fleet
Distributed installations, sensor nodes, and facility telemetry, with multi-site asset tracking and health monitoring.
Browse the catalogWhat you say next
- Add a warranty expiry date and a contractor reference to every unit.
- Split the sensor template so refrigeration and HVAC keep separate metrics.
- Alert the regional lead when any unit reports above 8 °C for ten minutes.
Each one is a schema change, a new record, or a new automation, applied to a live workspace in the same session.
Pricing
Start free. Upgrade when record volume, retention, or automation usage grows.
Free
Perfect for trying things out
Start Free- Entities 250
- Attributes 25
- History retention 7 days
- AI Credits 1,000
- Dashboards 1
- Automations 5
Team
For teams running live workflows
Start Free- Entities 15,000
- Attributes 250
- History retention 30 days
- AI Credits 15,000
- Dashboards 15
- Automations 50
Business
For larger datasets and longer review windows
Start Free- Entities 50,000
- Attributes 600
- History retention 90 days
- AI Credits 50,000
- Dashboards 30
- Automations 100
All plans include: Unlimited users • REST API • 8 languages • Google Sign-In
Enterprise Data Sovereignty & Custom Self-Hosted Licensing
Zero Vendor Lock-in: Run on our managed cloud today. If your regulatory, air-gap, or infrastructure policies ever require on-premise execution, custom self-hosted licensing and full database export are available on request.
Frequently asked questions
What counts as an entity? +
Does Omnismith support SOC 2 and ISO 27001 compliance requirements? +
Can I self-host Omnismith or run on-premises for regulatory data sovereignty? +
Can I upgrade or downgrade anytime? +
What happens if I hit my limit? +
Give your agent somewhere to build.
The free tier carries 250 entities, 25 attributes, and 1,000 AI credits, with the same history, metrics, and API as every other tier. No credit card required.
Questions? [email protected]