Sign in
Use the token your gateway admin gave you.
Workflows are the programs your apps and automations run. They come in bundles (.flow files): some ship with the gateway, others you import or publish from AbstractFlow.
| Name | What it does | Version | Source | Used by | Available to users | Actions |
|---|
Default workflow per app
When an app asks for “an agent” without naming a workflow, the gateway runs this one.
A connection is something of yours the gateway listens to: a mailbox, a feed, a webhook address. Flows and automations use it by name.
| Connection | Handled by | Account | Status | Last event | Actions |
|---|
A connection only fetches while an automation uses it. Pause it to stop fetching.
Instructions agents load when a task needs them: curated ones ship with the gateway, imported ones are yours to edit.
| Name | What it does | Version | Trust | Source | Actions |
|---|
Tool servers over the Model Context Protocol: Test runs the real handshake and lists their tools.
| Name | Transport | Status | Tools | Actions |
|---|
Local providers
Engines that run models on this computer, and their server connections.
abstractcore engines status --probe| Engine | Supported | Installed | Version | Running | Reachable | Base URL | Models | Actions |
|---|---|---|---|---|---|---|---|---|
| Probing engines... | ||||||||
Keys: r refresh; on a focused row: i install, o open download page.
Remote providers
Cloud accounts and OpenAI-compatible servers. Keys stay on the gateway; only fingerprints are shown.
Available Providers
Configured providers available to this Gateway principal. Use their provider ids in Flow nodes and Core capability defaults.
| Name | Provider ID | Type | Models | Status | Actions |
|---|
Multimodal Capabilities
Sign in to edit provider/model defaults for this Gateway runtime.
What the next message generates, with its route from Multimodal.
Gateway
How this gateway is running right now. Pausing stops new workflow steps; the console and connected apps keep answering.
Memory & GPU
Live truth from the host behind this gateway: RAM pressure, accelerator memory, GPU load. Refreshes every 5 seconds while this tab is open; a failed probe is named below, never blanked.
Models
Model runtimes on this host — resident in provider memory by default (default ≠ loaded: configured capability defaults sit behind the toggle until they are really in memory). Unload frees the memory (the next request pays the full load again); a locked model refuses unload until unlocked or forced; Estimate asks the host how much context actually fits.
| Modality | Provider | Model | Resident | Size | Context | Flags | Actions |
|---|
Session caches
Prompt (KV) caches the runtime minted per session. Clearing one only costs re-encoding the prompt on that session's next turn.
| Session | Model | Size | Tokens | Created | Actions |
|---|
Summoned entities
Persistent identities living on this gateway, each with its own memory and runtime. Talk to one, or manage its lifecycle, substrate, capabilities, and prompt.
| Name | Entity ID | Born | State | Actions |
|---|