AIR Platform

Tamper-evident audit. Runtime guardrails. Full trace observability. Local-first.

🎯 Live Demo · 🌐 airblackbox.ai · 📦 Main Repo

air-platform is the Docker Compose orchestration for the AIR Blackbox stack. One command starts the gateway, the GenAI-aware OTel collector, Jaeger tracing, and a MinIO prompt vault — a self-contained, local sandbox for recording, replaying, and auditing AI agent behavior.

The gateway ships as a prebuilt, signed image from GHCR (with SLSA build provenance). Nothing here phones home: the default collector pipeline exports to your local Jaeger only.

The EU AI Act enforcement date for high-risk AI systems is August 2, 2026. Companies deploying AI agents — tool-calling LLMs that act autonomously — face mandatory requirements around logging, transparency, human oversight, and data governance. Penalties reach €35M or 7% of global turnover.

Most compliance platforms target CISOs with top-down dashboards. AIR Blackbox gives developers the building blocks to make agents audit-ready by default.

git clone https://github.com/airblackbox/air-platform.git
cd air-platform
cp .env.example .env    # optional: set OPENAI_API_KEY, GATEWAY_KEY
make up                 # gateway + collector + Jaeger + MinIO
make status             # check health

Point your agent at the gateway instead of the provider — everything else in your code stays identical:

client = OpenAI(base_url="http://localhost:8080/v1")

Every LLM call now produces an HMAC-SHA256 chained, replayable audit record in ./runs/, plus an OpenTelemetry trace you can inspect in Jaeger at localhost:16686.

Ports

┌─────────────────────────────────────────────────────────────┐
│                       YOUR AI AGENTS                         │
│   (OpenAI SDK · LangChain · CrewAI · AutoGen · any client)  │
└──────────────────────────┬──────────────────────────────────┘
                           │ base_url swap
                           ▼
┌─────────────────────────────────────────────────────────────┐
│                    AIR GATEWAY  :8080                        │
│   OpenAI-compatible proxy · HMAC-SHA256 audit chain          │
│   guardrails (budgets, loop & retry protection, tool         │
│   allowlists) · PII / injection scanning · replay records    │
└───────┬──────────────────────────────┬──────────────────────┘
        │ OTLP traces                  │ .air.json records
        ▼                              ▼
┌───────────────────────────┐   ┌─────────────────────────────┐
│   OTEL COLLECTOR :4317    │   │   ./runs/  +  MinIO vault   │
│   normalize → prompt      │   │   (signed, replayable,      │
│   vault → redact/metrics  │   │    tamper-evident)          │
└───────────┬───────────────┘   └─────────────────────────────┘
            ▼
┌───────────────────────────┐
│      JAEGER UI :16686     │
└───────────────────────────┘

The collector pipeline (collector.yaml, editable) runs three AIR processors in order: genai_semantic_normalizer (vendor attrs → gen_ai.* conventions), promptvault (offload prompt/response bodies to encrypted storage, replace with refs), and genaisafe (PII redaction, token metrics, loop detection).

See the full compliance mapping for article-by-article details. AIR Blackbox checks technical requirements — it is a linter for AI governance, not a legal tool.

Framework trust layers for LangChain, CrewAI, OpenAI Agents SDK, Claude Agent SDK, AutoGen, Google ADK, and Haystack ship inside the main airblackbox repo (air_blackbox.trust.*).

The scanner runs against any Python AI project — no stack required:

pip install air-blackbox
air-blackbox comply --scan . -v     # gap analysis: Articles 9–15
air-blackbox replay                 # replay recorded episodes
air-blackbox evidence               # signed evidence bundle for auditors

Integration tests run against the live stack:

make up
make test    # 6 health/audit tests; +1 live LLM round-trip when OPENAI_API_KEY is set

If the stack isn't running, the suite skips with instructions instead of failing. CI boots the full compose stack and runs the same tests on every push.

All AIR Blackbox components are released under the Apache License 2.0.

We welcome contributions. See CONTRIBUTING.md for guidelines.

If AIR Blackbox is useful to you, a star helps others find it.

Questions or feedback? Start a Discussion.

AIR Blackbox — Agent Infrastructure Runtime
Compliance infrastructure for autonomous AI agents