Agent Orchestration Platform
Start here when you want the hosted NFLTR overview. This page explains how planners, remote workers, dashboard operators, timeline reads, managed capabilities, and secure routing fit together on nfltr.xyz.
Sign up on nfltr.xyz, create an API key, and install the nfltr CLI before you follow the workflow below.
What changes when NFLTR is the platform layer?
Coordinator Dispatch
Dispatch task contracts from the CLI (nfltr orch start task) or the dashboard's + Start form so the coordinator can match them to a connected worker through nfltr.xyz.
Worker Execution
Workers of different flavors (Claude Code, Codex, Cursor, or a generic MCP backend) connect to the fleet and execute matching contracts. Per-task dispatches carry task_mode contracts plus optional actor metadata.
Operator Visibility
The Causal Operations Atlas exposes an attention rail, causal canvas, task drawer, and Trace Lab alongside task-mode coverage, terminals, managed capabilities, and route inventory in one operator console.
Timeline Reads
Use nfltr orch task status --task TASK_ID --events 20 or nfltr orch task watch --task TASK_ID to inspect one compact task view instead of stitching separate APIs manually.
Recovery
Resume or re-dispatch nonterminal work with planner-local state and coordination bundles instead of minting duplicate tasks blindly.
Secure Transport
Verified routing, private shares, direct hostnames, browser terminals, and traffic capture still exist underneath the orchestration layer.
Core workflow
- Save an API key:
nfltr config add-api-key YOUR_KEY - Bring a worker fleet online with explicit capability labels — see Fleet Management
- Dispatch a task contract:
nfltr orch start task --task-mode implement --objective "..." --watch - Inspect one compact task view with
nfltr orch task status --task TASK_ID --events 20ornfltr orch task watch --task TASK_ID - Use the Causal Operations Atlas for operator readiness, run evidence, Trace Lab, managed surfaces, route inventory, and terminals
Where this is better than one local agent loop
| Need | Why NFLTR helps |
|---|---|
| Cross-machine execution | Workers can run on other machines and networks while the planner keeps one hosted control-plane entrypoint. |
| Capability-first dispatch | Workers advertise task-mode, tool, pool, workspace, and environment labels so the coordinator matches explicit task contracts instead of assuming fixed roles. |
| Operator inspection | The dashboard and timeline helpers give operators a concrete surface for readiness, history, and route state. |
| Recovery and handoff | Planner-local recovery and coordination bundles make resume or redispatch possible without ad hoc glue. |
Next reads
- Getting Started — sign up, install the CLI, and dispatch your first hosted task.
- AI Agent Orchestration — end-to-end planner-worker tool surface.
- Local Agent Orchestration — local-first operator walkthrough.
- Local Repo Improvement — isolated worker clones and git-backed patch return.
- Dashboard — Causal Operations Atlas (attention rail, causal canvas, task drawer, Trace Lab) plus readiness, terminals, traffic, and managed surfaces.
- Dashboard walkthrough — Atlas operator workspace targets and CLI cross-checks.
- When to use NFLTR — hosted
nfltr.xyzversus production standalone GCE VM (embeddedweb/); Kubernetes/Helm/make deploy-uiare legacy/non-production.
Use NFLTR as the orchestration layer
Start with the hosted planner entrypoint on nfltr.xyz, bring workers online, then use the Causal Operations Atlas and timeline tools for the operator view.
Read Orchestration Guide Getting Started Open Dashboard Homepage →