> ## Documentation Index
> Fetch the complete documentation index at: https://docs.gdilabs.io/llms.txt
> Use this file to discover all available pages before exploring further.

# Worker

> The orchestrator. A Python LangGraph service that consumes the job queue, augments prompts with retrieved context, runs every paid job through the scenario runtime, dispatches to free or paid models, and streams typed events back through...

## What it does

* Pulls jobs from Redis.
* Runs the context engine before routing — pulls knowledge-hub augments and RAG-retrieved snippets into the prompt.
* Routes paid jobs through `ScenarioOrchestrator`; see [Scenarios runtime](/developers/components/scenarios-runtime) for the engine and [Scenario execution](/how-it-works/scenario-execution) for the per-level walkthrough.
* Tracks escalation state and per-conflict retry counts.
* Emits typed `AgentEvent` JSON (including the Phase-4 `level_*`, `verifier_*`, `gate_decision`, `budget_*` events) to a Redis list and pubsub channel.
* Persists status, cost, and audit trail.
* Provisions per-project workspaces (clone, scaffold, trigger Vercel deploy).

## Model dispatch

Free and paid model dispatch with health-ranked provider fallback.

* **Free providers**: Ollama (local or remote), including thinking models.
* **Paid providers**: Claude and OpenAI-compatible endpoints.

A failing provider is de-preferred for subsequent jobs until its health recovers.

## Hierarchy invariants

* Sub-job nesting depth is capped at 2.
* Per-conflict escalation count maxes at 3 retries before the conflict is escalated upward.
* The L4 executor wraps token streaming with periodic heartbeats so long generations stay alive end-to-end.
* Knowledge-hub edits hot-reload into running workers via a refresher thread — no worker restart needed.