Arbiter Platform · Hosted workspace

The workspace your agents run in.

arbiter.run is the hosted home for supervised agent workflows — a dashboard for everything in flight, a visual agent builder, and a navigable memory graph, all on the open runtime and the same HTTP API.

01Same surface

The API the CLI speaks

The workspace runs on the exact HTTP API the open runtime exposes. What you use hosted, you can run yourself.

02Billing

Credits, keys included

Top up credits and run — provider keys included and metered per request. No separate provider accounts to wire up.

03Isolation

Per-tenant by default

Agents, conversations, memory, and artifacts are scoped to your workspace. Bearer-token auth, isolated per tenant.

04Live

Every turn, streamed

Watch execution as it happens — sub-agent fan-outs, tool calls, advisor verdicts — over the same SSE stream a client would read.

Workspace

Every surface onto the same runtime.

Each view is a window onto the same orchestration API — and onto the same agents, memory, and runs.

Dashboard

A control center for everything in flight.

Land in the workspace and see what your agents are doing — recent tasks, scheduled runs, a live activity feed, and a window onto memory, all on one screen.

  • Recent tasks across your agents, with status at a glance.
  • Scheduled runs and what fires next.
  • A live activity feed and a window onto the memory graph.
The Arbiter dashboard — recent tasks, a schedules panel, an activity monitor, and a memory graph panel.
Agent builder

Compose an agent on a canvas.

Wire up an agent visually — its identity and goal, the rules it follows, the capabilities it can call, the MCP servers it reaches — and see how the pieces connect before you publish.

  • Identity, generation settings, and advisor mode in one panel.
  • Capabilities and MCP tool servers wired in as nodes.
  • Draft, publish, and reuse agents across the workspace.
The Arbiter agent builder — a node canvas connecting rules, identity, capabilities, and MCP servers for an agent named Strategist.
Memory

A memory graph you can actually see.

Memory is a typed graph of entries and relations, not a flat vector dump. Browse it, filter by type, and follow how one memory connects to the next.

  • Typed entries — user, project, reference, context, lessons.
  • Relations rendered as a navigable graph.
  • Inspect any node and trace what it links to.
The Arbiter memory view — a force-directed graph of memory entries with a type filter sidebar and a detail panel.
Closed beta

Get a workspace.

arbiter.run is in invite-only beta. Credit-based pricing — top up, run, stop when you want. No telemetry on your prompts, ever.