Positioning

The local-first graph OS — not another agent workflow builder

A lot of products called “AI OS” are really orchestration UIs: triggers, pipelines, RAG nodes, MCP deploy buttons — Langflow, n8n, or enterprise data engines with a visual builder on top. Trellis is a different layer: a causal graph you own where notes, code, agents, and decisions are entities in one substrate. Studio is a projection over that graph — not a pipeline canvas.

Run Studio locally Read the vision

$npx trellis studio

Opens Trellis Studio in your browser. Local-first by default.

Three layers people confuse

Layer	Examples	What it optimizes for
Datacenter “AI OS”	VAST, GPU fabrics, exabyte storage	Enterprise inference at scale
Workflow / agent builder	Langflow, n8n, “deploy as MCP tool” UIs	Composable pipelines over existing data silos
Graph OS (Trellis)	trellis.computer	One owned graph: projections instead of apps, append-only history, agents as signed peers

What Trellis actually is

Causal stream — every change is an immutable op with deterministic ordering and replay.
Projections, not apps — board, table, timeline, and graph are views over the same entities.
Sovereignty — Ed25519 identity, .trellis files on disk, offline by default.
Agents as peers — signed ops, decision traces with rationale and alternatives — not a black-box side channel.

Pipelines glue services together. Trellis replaces the silo: the graph is the integration layer.

Not the other Trellis

Healthcare admin, 3D asset models, and legal research tools also use the name Trellis. If you landed here from a confused search, see which Trellis? .