Rhetra Get in touch
Platform and implementation, one team

Make your operation executable.

A mine, a hospital, a fleet, a two-person shop. Rhetra reconciles your systems into one typed model of what you actually run, then gives every workflow, application and agent the same governed way to change it.

01  Connect

Connect everything you already run.

Batch syncs, live streams, file drops and agent writes, read continuously and typed on the way in.

from any application
FORTNOXSAPCRMMES / SCADAWMSHRMBI TOOLS+550 via Airbyte
from files, docs, and notes
CSVPDFSPREADSHEETSDOCSNOTESJSONXML / EDI+37 more
from emails, chats, and forms
EMAILWHATSAPPSMSWEB FORMSMEETINGSHELPDESK+24 more
from physical sensors
TEMPERATUREFLOWRPMVIBRATIONGPS+12 more
 
Everything arrives here
from offline sources
SCANSPAPER FORMSHANDWRITINGLABELS+3 more
from databases / warehouses
POSTGRESSQL SERVERLAKEHOUSEOLAPS3 / OBJECT STOREGRAPH+5 more
from streams and the web
WEBSOCKETSWEBHOOKSPOLLINGPUBLIC PAGES+9 more
from agents and apps
MCP CLIENTSAI PROPOSALSSDK APPSvia actions
All sources supportedBatch syncsLive streamsFile dropsAgent writes
02 The ontology
THE SEMANTIC LAYER · OBJECTS, PROPERTIES & LINKS

Describe the operation once.

Ingested records become typed objects. Customers, orders, invoices and sensors, each carrying properties, real relationships to one another, and the actions that are allowed to change them.

The schema is defined at runtime. Add a property or link two types and every record follows within the hour, without a migration.

Every change is recorded on two time axes. Ask what was true last quarter, or what you believed last quarter. Both are answerable.

TYPES · LINKS · ACTIONS · FULL HISTORY
03 Actions & functions
THE KINETIC LAYER

Write logic against the model.

The model carries the levers that move the operation. Logic is written against the objects you defined, in server-side TypeScript, with a client generated from your ontology: objectSet(), traversal, aggregation, all typed.

A typed client, generated

Rename a property and the function that uses it fails to compile. The ontology is the contract.

LLM steps, tested like code

Logic chains mix LLM calls with conditionals and loops. Every chain ships with an eval suite, so a prompt change reports a pass rate before it reaches production.

Read-only by design

Functions compute. They don’t commit. When logic needs to change something it calls an action, and inherits its rules.

Functions read the ontology freely. Every write routes through an action, and its criteria.

04 Operate

The model runs the operation.

Every write goes through an action with submission criteria, so the rules live on the type rather than in someone’s head. Workflows chain actions into processes, automations watch the data and react, and decisions write back to the ontology and out to the source systems the records came from.

Actions

The only write path. Submission criteria decide who can change what, enforced on dispatch. Same rules for a person, a workflow, or an AI agent.

Workflows

Multi-step processes with branching, waits, and checkpoints. A checkpoint pauses the run until a reviewer approves in the Inbox.

Automations

React to the data itself: entity created, entity updated, or a schedule. Conditions, retries, and cooldowns keep reactions deliberate.

AND YOUR OWN APPS ON TOP

One command generates a fully typed TypeScript client from your ontology. Build real apps on it. The platform hosts them, wired to the same actions and permissions as everything else.

OR ASSEMBLE THEM

Workshop builds operational UIs without writing code. Tables, detail views, investigation graphs, action forms.

05 AI

Agents operate the model.

Rhetra speaks MCP. Point Claude, or any MCP client, at a workspace and it receives the full toolset: search, traverse, aggregate, run pipelines, trigger actions. An agent gets the same toolset an operator gets, and the same limits on what it can commit.

A live MCP session. Reads run free, writes route through an action, ontology edits become proposals.

Watch a real session, prompt to app →

One command

rhetra-sdk mcp connects any MCP client to your workspaces with a personal token.

MCP servers from your types

Every registered app can expose its slice of the ontology as a hosted MCP server. The tools are generated from your types and actions, not hand-written.

Guardrails included

Agent writes go through actions and their submission criteria. Ontology edits land as proposals you approve in the Inbox. The rules apply to an agent exactly as they apply to a person.

Get in touch

Tell us what you need to run.

Based in Sweden. Available remotely.

Email us · Johan-sandgren@hotmail.com