> ## Documentation Index
> Fetch the complete documentation index at: https://docs.orq.ai/llms.txt
> Use this file to discover all available pages before exploring further.
# The Orq Flow
> A unified workflow for teams to manage the agent lifecycle in a central platform, from planning requirements to operating in production.
**orq.ai** organizes AI development into five connected stages. Each stage has a clear owner, feeds into the next, and the later stages loop back, creating a continuous improvement cycle for your AI applications.
***
## Plan
Product Manager
The most important thing to define upfront is what **quality means for your feature**. LLM outputs are probabilistic: there is no single right answer, so success criteria must be concrete enough to measure. The quality bar you set here becomes the foundation for the evaluators you write, the datasets you curate, and the experiments you run both before and after production.
**What you'll do:**
* Define the intended behavior and the user interactions the feature must handle
* Specify measurable quality criteria (tone, accuracy, format, safety, etc.)
* Scope the work into [Projects](/docs/projects/overview) and invite [Members](/docs/administer/permissions/overview) with the right roles
**Outcome:** A shared definition of quality that every downstream stage can be measured against.
***
## Build
Full Stack Engineer Subject Matter Expert
Engineers implement the application scaffolding; subject matter experts own the natural language layer: the system instructions, prompt structure, and retrieval configuration that drive model behavior.
**What you'll do:**
* Author and iterate on prompts in the [Playground](/docs/playground/overview)
* Configure [Agents](/docs/agents/build) for multi-step workflows
* Connect [Knowledge Bases](/docs/knowledge/overview) for retrieval-augmented generation
* Version prompt configurations in [AI Studio](/docs/prompts/overview) independently of application code
**Outcome:** A versioned, testable configuration ready for evaluation.
***
## Experiment
AI Engineer Data Scientist
Validate configuration quality at scale before shipping. Run a representative input set against one or more configurations and score each output against the quality criteria defined in Plan.
**What you'll do:**
* Curate representative [Datasets](/docs/datasets/overview) of real or synthetic inputs
* Run configurations side by side in [Experiments](/docs/experiments/overview)
* Score outputs automatically with [Evaluators](/docs/evaluators/build) (LLM-as-a-Judge, Python, or human review)
**Outcome:** Evidence-backed confidence that the configuration meets the quality bar before it reaches production.
***
## Deploy
AI Engineer Full Stack Engineer
Publish a validated configuration to production. Model, instructions, and parameters can be updated from the Studio without a code change or redeployment, keeping iteration cycles short once the feature is live.
**What you'll do:**
* Publish prompt and agent configurations as versioned [Deployments](/docs/deployments/overview)
* Route traffic across models with fallbacks, load balancing, and cost controls via the [AI Router](/docs/router/using-the-router)
**Outcome:** A live, versioned feature with no redeployment required to iterate on model behavior.
***
## Operate
AI Engineer Data Scientist
Every request is captured automatically. Use production signal to understand real behavior, catch regressions, and continuously improve, closing the loop back into Experiment.
**What you'll do:**
* Inspect requests step by step in [Traces](/docs/observability/traces)
* Monitor agent-level performance and step timelines in [Control Tower](/docs/control-tower/getting-started)
* Label production spans with [Annotations](/docs/annotations/build) to build evaluation datasets from real traffic
* Track cost, error rates, and model usage trends in [Analytics](/docs/analytics/dashboards)
**Outcome:** A continuous feedback loop that turns production traffic into the next round of improvements.