Open-source asset context platform
One asset context. Every decision.
One shared, live picture of every asset — for your people, your apps, and your AI agents. Your data never moves; henosis connects what you already have.
$ uv tool install henosisOpen-source · self-hosted · MIT · Python 3.11+
One opinionated model
henosis's ontology isn't a blank graph of entities and relationships. Whatever you're reasoning about — a well or a pump, but equally a maintenance schedule, a work order, or an event — a decision about it draws on the same few things:
Identity
What is it?
a well, a pump, a work order
Location
Where is it?
a wellhead at 27.7°S, 140.3°E
Observation
What's it doing, now and over time?
gas rate over the last 30 days
Documents
What do we know about it?
the P&ID, the last inspection report
Relationships
What's it connected to?
the pipeline it feeds, its open defects
Decision → Action
act on the asset — its new state becomes context for the next decision
a closed feedback loopBecause these are first-class roles — not generic nodes and edges — henosis knows an observation is a time-series and a location is a place, and drives maps, trend charts, and agent tools straight from your model. A generic ontology is a blank canvas; this isn't.
Focus beats a bigger graph
Ask both the same question. A generic knowledge graph is an undifferentiated blob — an agent wades through it, touching node after node, spending context and time to reach an answer. henosis reads the same connections, but its first-class roles tell it exactly which few to follow — so it walks straight to the same answer on a fraction of the context. And that one focused model is reachable by people, apps, and agents alike.
Generic knowledge graph
touches everything to find the answer
Only reachable by
Agent
MCP tools
henosis
same graph — the roles route straight to the answer
- Identity
- Location
- Observation
- Documents
- Relationships
Reachable by
Human
Canvas
App
REST API
Agent
MCP tools
Why henosis
Universal
One model, used the same way by a person, an app, or an AI agent — through the Canvas, a REST API, or MCP tools. Most tools serve just one of the three.
Open
Open-source and self-hosted under MIT — run it yourself, read every line, no lock-in. The rest of this lane is closed and expensive.
Federated
Your data stays where it already lives — the historian, the maintenance system, the map. henosis resolves it at call time; it never copies or moves it. No ETL, no migration.
Actionable
It closes the loop — your people and your agents act on what they find, back on the source systems, instead of stopping at read-only answers.
Works with what you already run
Federation means henosis reads from your systems at call time and never copies or moves the data. Point it at what you already run:
Historian & signals
- AVEVA PI (OSIsoft)
- REST / HTTP APIs
Databases
- PostgreSQL
- MySQL
- Snowflake
- SQLite
Files & lakes
- Parquet
- CSV
Geospatial services
- ArcGIS
- WFS
Documents
- Document stores
Not on the list? A connector is a class you register — no fork, no core change.
Watch the agent drive the real Canvas
From a constrained-wells question to a work order written back — the agent finds the choke point and proposes the fix. Every clip is the real henosis Canvas, not a mockup.
Good morning.
Demo data is synthetic, or derived from publicly available sources — the South Australian PEPS-SA petroleum register and Geoscience Australia. No operator-confidential data is used.
The same ontology, from Claude Desktop
One unedited session over the Cooper Basin example — an open question, down through the network to the defect behind it, a work order raised, and the whole thread handed back as a forwardable brief. Jump to any beat.
Get running in three commands
henosis is a CLI tool (Python 3.11+). Install it, then compile and serve the worked example — a synthetic upstream gas operation that ships with the repo.
$ uv tool install henosis# Scaffold a worked example, then seed, compile and sync it
$ henosis init cooper-basin# Serve the REST API (/v1) + explorer, and mount MCP at /mcp
$ henosis serve -p cooper-basin --mcpFull instructions and the grammar are on GitHub and in the Docs.