FlowX Observatory is the operational layer for AI agents. It turns raw LLM telemetry into the four things you need to run agents in production: visibility, control, regulatory evidence, and a credible ROI story.Documentation Index
Fetch the complete documentation index at: https://docs.flowx.ai/llms.txt
Use this file to discover all available pages before exploring further.
What problem this solves
Generic LLM dashboards stop where the hard work starts. Observatory covers the full path: from the first trace your SDK emits to the EU AI Act control your auditor wants signed off, and the Net Present Value your CFO wants to see. Reach for Observatory when you need to:- See exactly what every agent, chain, and tool call did
- Enforce policies before unsafe behaviour reaches users
- Stay continuously aligned with EU AI Act, NIST AI RMF, or ISO 42001
- Quantify the financial impact of AI investments
The four pillars
Observability
Real-time tracing, cost and latency analytics, drift detection, and threshold alerts for every LLM call.
Governance
Policy engine, risk scoring, evidence collection, and assessments to keep agents accountable.
Compliance
EU AI Act, NIST AI RMF, and ISO 42001 controls mapped to your runtime with automated evaluation.
ROI & value
Risk-adjusted financial ROI, payback, NPV, and Monte Carlo sensitivity per agent and per project.
How the pieces fit together
The SDK decorates your agents, chains, and tools and streams events to the Observatory API. The API stores telemetry in PostgreSQL with thepgvector extension and optionally forwards content through Observatory Guards for safety checks. Every pillar of the UI reads from the same data — what you observe is what you govern, certify, and report on.
Concept glossary
| Term | What it means in Observatory |
|---|---|
| Org | The top-level tenant. Owns apps, datasets, prompts, and experiments. |
| App | A single AI app. Owns runs and API keys. |
| Run | One execution trace of an agent, chain, or tool. |
| Event | A log entry inside a run — for example chain_start, tool_end, chat. |
| Dataset | A test set of inputs used for evaluations. |
| Prompt | A versioned prompt template. |
| Experiment | An evaluation run that scores a prompt or model against a dataset. |
| Policy | A governance rule evaluated against runs. |
| Evidence | An artefact that proves a control is met. |
| Risk score | A six-dimensional score per app. |
Where to go next
Getting started
Connect your first project and emit your first trace in 15 minutes.
SDK reference
Install the SDK and learn the
@agent, @chain, and @tool decorators.Self-hosted setup
Stand up Observatory with Docker Compose for development or on-premise deployments.
AI Platform
Build the agents that Observatory observes.