The company
Linea Labs
A small team building the substrate production agents need: durable execution, memory, human-in-the-loop, and the fix loop that sits on the same data model.
About
We are building the runtime and control plane for production agents.
You ship the agent. Linea executes the graph and keeps the records around the run. Workflow stays the name of that graph, not the name of the product.
The product
Linea
An agentic runtime and control plane. Author the graph, run it on the host, and operate the agent from the records that execution wrote.
We execute the graph
Not observe it, not orchestrate a call out to something else. Because Linea runs the workflow, it holds the checkpoint, the signal, and the case you freeze from a failure.
The fix loop is on the host
Replay the failing step against the input it saw. Publish a version. The suite checks the fix. An observability tool can show the break; it never ran the graph.
One graph, two surfaces
The builder and the SDK emit the same workflow definition. An agent may use several graphs. We will not ship a second product for a second editor.
The product
Runtime, replay, and memory.
Runtime
The host that runs the agent graph.
Author the graph on the canvas or in TypeScript. Start an execution from your app, a webhook, or a schedule. The host writes a checkpoint after every node.
Replay
Replay the step that failed.
An observability tool can show the failure. It never ran the graph, so it has no checkpoint to replay from. Linea does.
Memory
Memory that survives the run.
The graph asks for memory. The host keeps it. remember, recall, and search are calls on the run — not a weekend of schema design.