openai-radar

OpenAI infrastructure FinOps SDK — part of the Hyperscaler Radar suite.

Scan your OpenAI org for assistants, vector stores, fine-tunes, batch jobs, and token usage. Detect external service relationships in assistant instructions, flag cost anomalies via a findings engine, and export inventory to CSV or a draw.io architecture diagram — all through a clean Python SDK that mirrors the openai-agents Runner API.

Install

pip install openai-radar                   # SDK + CLI + CSV and draw.io export
pip install openai-radar[agents]           # + openai-agents integration
pip install openai-radar[all]              # everything

Requires Python 3.10+.

CSV and draw.io export are part of the base install — no extras needed, same as the rest of the Radar suite. The [csv] and [drawio] extras still resolve (as no-ops) so older pins keep working.

Quick-start

from openai_radar import RadarClient, Runner

# Reads OPENAI_API_KEY from environment
client = RadarClient()

result = Runner.run_sync(client)

print(result.summary())
result.export_csv("./out/")
result.export_drawio("./out/openai_arch.drawio")

With admin key (org-wide visibility)

client = RadarClient(
    api_key="sk-proj-...",
    admin_key="sk-admin-...",   # unlocks cross-project usage data
)
result = Runner.run_sync(client, config)

Scoped to a project

from openai_radar import RadarClient, Runner, RunConfig

client = RadarClient()
config = RunConfig(project_id="proj_xxx", usage_lookback_days=7)
result = Runner.run_sync(client, config)

CLI

# Full scan — findings table to stdout
openai-radar run

# CSVs + draw.io diagram
openai-radar run --csv-dir ./out --drawio-file arch.drawio

# JSON instead of a table
openai-radar run --output json --out-file scan.json

# Admin key for org-wide usage data
openai-radar run --admin-key sk-admin-... --csv-dir ./out

# Findings only
openai-radar findings

# Scope to a project, 7-day lookback
openai-radar run --project proj_xxx --lookback 7

# Print the version
openai-radar version

Flags follow the Radar suite convention: --output/-o selects table or json, --out-file writes the JSON payload, --csv-dir writes per-resource CSVs.

openai-agents integration

pip install openai-radar[agents]

from openai_radar.agents import build_radar_agent
from agents import Runner

agent  = build_radar_agent()
result = Runner.run_sync(agent, "Scan my org and flag any cost anomalies")
print(result.final_output)

Or compose individual Radar tools into your own agent:

from agents import Agent
from openai_radar.agents.tools import scan_assistants, run_findings, export_drawio

agent = Agent(
    name="My FinOps Agent",
    instructions="...",
    tools=[scan_assistants, run_findings, export_drawio],
)

Findings engine

Service relationship detection

AssistantScanner inspects each assistant's instructions field for external service signals and emits ServiceRelationship objects (visualized as edges in the draw.io diagram):

SDK structure

src/openai_radar/
├── client.py            # RadarClient (auth, project vs admin key)
├── runner.py            # Runner, RunConfig, RunResult
├── findings.py          # FindingEngine, Finding, Severity
├── models/base.py       # Pydantic v2 models for all resource types
├── scanners/            # One scanner per resource type
├── exporters/           # CSV + draw.io exporters
├── agents/              # openai-agents tools + build_radar_agent()
└── cli.py               # openai-radar CLI

Part of the Hyperscaler Radar suite

aws-radar · gcp-radar · azure-radar · oci-radar · openai-radar · datadog-radar · coreweave-radar