Zero dependencies. Govern any AI agent. — Try the Demo
ATOMIC PILOT™

Wrap Any AI Agent.
Govern Everything.

Atomic Pilot is a zero-dependency Python SDK that wraps any AI agent with structural governance. Trust scoring, provenance chains, 5 geometric invariants, and cryptographic audit trails. One import. One wrapper. Full governance.

Get Early Access Try the Demo
govern_my_agent.py
from atomic_pilot import Pilot

# Wrap any AI agent in 3 lines
pilot = Pilot(agents=[
  "analyst", "writer",
  "reviewer", "deployer", "auditor"
])

# Every decision is now governed
result = pilot.decide(
  agent="analyst",
  action="approve_model",
  context={"model": "gpt-4o"}
)

# Conservation law: K = 7.0
assert result.K == 7.0 # Always
assert result.chain_hash # Provenance
assert result.invariants_passed # All 5
0
Dependencies
355
Tests Passing
91.6%
Coverage
v1.0
Stable
100%
TrojAI Score

Three Steps to Governed AI

1

Wrap

Import Atomic Pilot and wrap your agents. Any Python agent, any framework, any LLM provider.

2

Govern

Every decision passes through structural enforcement. Trust scores update. Provenance chain extends.

3

Prove

Cryptographic receipts for every action. 5 geometric invariants checked. Audit-ready from the first call.

Interactive SDK Console

Type commands or click the examples below. This simulates the Atomic Pilot SDK running in a Python REPL.

python3 — atomic-pilot v1.0
>>>

What's Inside

Conservation DAG

Directed acyclic graph of trust transitions. Every state satisfies K = 7.0. Anomaly detection built in.

BLAKE3 Provenance

Cryptographic hash chain with Merkle trees. Every decision has a tamper-proof receipt. Optional BLAKE3 (falls back to SHA-256).

5 Invariant Checker

Velado Bound, Excess Action, Byzantine Tolerance, Phase Boundary, Temporal Coherence. All 5 checked every transition.

Audit Engine

8-check governance audit with sealed bundles. Generate compliance evidence packages on demand.

Multi-LLM Tribunal

Multiple LLM judges with Byzantine filtering. Consensus-based decisions. Eliminates single-model failure.

Phase Detection

5-phase model (crystalline, solid, liquid, gas, plasma). Automatic detection of trust regime transitions.

Δ

Govern your AI in 3 lines of code

Zero dependencies. Full structural governance. Instant provenance.