Sixb: the operating layer for enterprise AI
Today, we are releasing Sixb, an open-source TypeScript framework for building operational software used by humans and AI agents.
Where we come from
We spent the last few years in the building industry, creating digital twins to improve building efficiency and operations.
That experience left us with a clear lesson: the digital twin itself is not the product. The value comes from use cases that solve real problems. As new needs emerge, each use case adds to the operational model. Together, these small wins compound into a connected system grounded in real operational needs.
One use case
A clear operational need
One application
Value in a real workflow
A complete operational layer
Expanded as new needs emerge
Build the system one use case at a time.
The problem
Today, almost every company uses an AI assistant: ChatGPT, Claude, or Gemini.
But it remains an assistant.
It answers questions when prompted, often with incomplete context. It does not know what changed in the business, which rules apply, what users are allowed to see, or which actions require approval.
Companies also send large amounts of business data to AI providers because better context usually produces better answers. Sensitive data, permissions, and governance are often considered too late.
The model should not contain the company’s operations. It should be one component connected to them.
Our thesis
AI is making interface generation and glue code cheaper.
The difficult part is no longer creating another form, dashboard, or isolated internal tool. It is modeling the business context, rules, permissions, and workflows behind them.
MCP solves access to tools, but it does not solve business context, permissions, or governance.
Generating an app with an AI tool can solve an isolated need quickly, but it often creates another disconnected application with its own data, logic, and permissions.
You cannot get a useful AI implementation off the shelf. Every company operates differently.
What Sixb does
Sixb lets you model the operational layer of an organization so teams and AI agents can work in the same environment.
Connect your data
CRM, ERP, file systems, meeting transcripts, equipment, or internal APIs.
Model your business
Customers, projects, equipment, quotes, contracts, and the relationships between them.
Define your rules
A customer must receive a quote within two days. A contract determines whether a repair is covered.
Define permissions
Decide what each user or AI agent can see and which actions they can perform.
Define workflows
React to business events, request human approval, and execute actions in source systems.
Build custom applications
Give teams and AI agents the same operational context and controlled actions.
AI in the enterprise should not only answer questions. It should be able to operate safely as part of the team.
A concrete example
The repository includes Northline Operations, a fictional commercial service company built entirely with Sixb.
When an equipment alarm is received, Sixb can:
- 01
Create a service case.
- 02
Retrieve the customer, facility, equipment, and active contract.
- 03
Apply coverage and response-time rules.
- 04
Recommend an available and qualified technician.
- 05
Ask a dispatcher to approve the assignment.
- 06
Create the work order in the field-service system.
- 07
Track the diagnosis, quote, repair, and recovery.
The application, workflows, rules, actions, and source systems all run locally with deterministic example data.
How Sixb works
Model
You define an ontology describing the business domain:
- Object types
- Properties
- Telemetry
- Relationships
An ontology works like TypeScript types: it defines the contracts that your business objects must respect.
You also define the actions that can be executed:
- Create a quote
- Dispatch a work order
- Generate a meeting checklist
- Update a customer record
import { LocalBlobStorage } from "@sixb/blob-local"import { createSixb, InMemoryBroker, InMemoryQueues } from "@sixb/core"import { DuckLakeStorage } from "@sixb/ducklake"import { SqliteStorage } from "@sixb/sqlite"export const sixb = createSixb({ id: "service-operations", broker: new InMemoryBroker(), storage: new SqliteStorage({ path: ".sixb" }), lakeStorage: new DuckLakeStorage({ catalog: { type: "duckdb", path: ".sixb/lake/catalog.ducklake" }, dataPath: ".sixb/lake/data", }), blobStorage: new LocalBlobStorage({ basePath: ".sixb/blobs" }), queues: new InMemoryQueues(),})Connect and transform
Connectors communicate with source systems.
Syncs ingest raw data into datasets stored in S3-compatible object storage.
Pipelines clean and transform those datasets using SQL or TypeScript.
Projections turn the resulting data into typed objects, relationships, and telemetry defined by your ontology.
Govern
Rules evaluate the operational state of the business.
Permissions control what users and AI agents can see and do.
Workflows coordinate actions, events, schedules, and human interventions.
Build and execute
Once the foundations are in place, you can build the software your operations actually need:
- A dashboard combining data from multiple systems
- A human-supervision interface for AI actions
- An operational application adapted to a specific team
- An AI assistant that understands the business without accessing information it should not see
- Event-driven agents that react without waiting for a user prompt
What is available in 0.1.0
| Area | Available today |
|---|---|
| Modeling | Ontologies, typed objects, properties, telemetry, links, CRUD, delete, and restore |
| Queries | Typed queries, counts, facets, pagination, and graph expansion |
| Data | Connectors, datasets, syncs, pipelines, and projections |
| Automation | Actions, rules, schedules, workflows, and human interventions |
| Agents | Tool use, sandboxed execution, and live run streaming |
| Events | Durable domain events, WebSocket streams, filtering, and run logs |
| Security | Authentication strategies, groups, roles, policies, and grants |
| Files | Direct uploads, multipart uploads, and downloads |
| Storage | PostgreSQL, SQLite, in-memory storage, and DuckLake |
| Applications | Atlas admin console, custom apps, and a typed React client |
The complete roadmap and technical limitations are available in the repository.
Current limitations
Sixb is still early. Expect breaking changes as the framework evolves.
Contributing
We are looking for developers willing to:
- Run the Northline example
- Challenge the architecture
- Build their own operational application
- Tell us what feels missing or over-engineered
- Contribute connectors, storage providers, examples, and core features
Sixb is at the beginning. We would love to see what you build with it.
Explore Sixb on GitHubWritten by
