AGENT
H.A.L.O.
Your agents, sovereign.
Wrap any AI agent with cryptographic identity, post-quantum comms, and tamper-proof observability. All local. Zero telemetry. Every operation backed by machine-checked proofs.
curl -fsSL https://raw.githubusercontent.com/Abraxas1010/agenthalo/master/install.sh | bashTamper-Proof Recording
Wrap any AI agent CLI. Record every event — thoughts, tool calls, file edits, token counts, costs — into tamper-proof NucleusDB traces.
Cryptographic Identity
Each agent gets a DID from a genesis seed ceremony. Ed25519 + ML-DSA-65 dual signatures. Append-only identity ledger.
5-Layer Sovereign Comms
Privacy Controller → Nym mixnet → libp2p mesh → DIDComm v2 + A2A → ZK capability tokens. Hybrid KEM encrypted. Fail-closed by default.
They sandbox. We prove.
Every competing agent infrastructure solves containment — keep the agent in a box. AgentHALO solves trust: prove mathematically that the agent's identity is genuine, its data is intact, its communications are authentic, and its history is tamper-evident.
Runtime-to-proof provenance. Lean 4 theorems with zero sorry. Mirror modules link runtime to spec. Gate rejects unprovenanced operations.
Post-quantum cryptographic identity from genesis. NIST FIPS 203/204 algorithms protecting against future quantum adversaries.
Every NucleusDB mutation generates verifiable cryptographic evidence. Three commitment backends. Append-only monotone seal chain.
No central authority. Entropy-derived DID + EVM wallet. The agent owns its identity from the moment of genesis.
Groth16/BN254 zero-knowledge proofs. Agents prove authorization without revealing identity or grantor.
Category-theoretic composability. Properties proved in one subsystem compose naturally with all others via functors.
Integrated AgentPMT: USDC wallets on Base L2, budget-controlled spending, dynamic tool marketplace. Agents never touch private keys.
Every subsystem, documented.
Each module has its own dedicated page with architecture diagrams, code samples, and worked examples.
Verifiable database with SQL, vector search, and three commitment backends
Tamper-proof session recording for Claude Code, Codex, and Gemini CLI
Genesis seed ceremony, DID derivation, and five-key-pair hierarchy
Full post-quantum cryptographic stack with honest vulnerability disclosure
9-page command center with cockpit terminal, deploy catalog, and orchestrator
128+ native tools for database, orchestration, identity, trust, and containers
22 machine-checked Lean 4 theorems bound to runtime operations
Wraps, not replaces.
AgentHALO provides a trust container around any existing agent or framework. Bring your models, your tools, your chains.
Autonomous Agent Payments
Programmable USDC wallets on Base L2. Budget-controlled spending with vendor allow-lists, cadence limits, and mobile operator authorization. Agents never touch private keys.
MIT licensed. Full Rust source, Lean 4 proofs, and dashboard on GitHub.
The serious back.
Architecture, proofs, cryptography, and governance — expand any section to go deep.