BentoML

Enumerate and exploit BentoML model serving services.

Overview

The bentoml module targets BentoML services that expose REST APIs for model inference. It discovers service metadata, parses OpenAPI specs for prediction endpoints, tests inference access, and extracts Prometheus metrics.

Subcommands

Read-Only (no --force-exploit required)

Subcommand	Description
enum	Enumerate service metadata, health, and API routes
routes	Parse OpenAPI spec for all prediction endpoints with input schemas
metrics	Retrieve Prometheus metrics (request counts, latency, model performance)

Gated (requires --force-exploit)

Subcommand	Description
predict	Send a prediction request to test inference access

Flags

Flag	Required	Description
--target	Yes	BentoML service URL (default port 3000)
--header	No	Custom HTTP headers. Repeatable.
--endpoint	For predict	Prediction endpoint path (default: /)
--payload	For predict	JSON payload for prediction

Examples

# Enumerate service metadata
./aipostex bentoml --target http://127.0.0.1:3000 enum

# List all prediction routes from OpenAPI spec
./aipostex bentoml --target http://127.0.0.1:3000 routes

# Extract Prometheus metrics
./aipostex bentoml --target http://127.0.0.1:3000 metrics

# Test inference access (gated)
./aipostex bentoml --target http://127.0.0.1:3000 predict \
  --endpoint /predict --payload '{"input": "test"}' --force-exploit

Key Endpoints

Endpoint	Method	Purpose
/	GET	Service metadata (name, version)
/healthz	GET	Health check
/docs.json	GET	OpenAPI specification
/metrics	GET	Prometheus metrics
/<endpoint>	POST	Prediction endpoints (from OpenAPI spec)

Workflow Progression

discover network (discovers BentoML on :3000)
  -> bentoml enum (service metadata, routes)
    -> bentoml routes (detailed endpoint discovery)
    -> bentoml metrics (operational data)
    -> bentoml predict --endpoint <route> (inference test, gated)