nxd 0.3.2

Encrypted compute layer for AI agents

NXD

NXD is an encrypted compute layer for AI agents. It wraps fully homomorphic encryption, credential vaulting, and privacy primitives behind a single Python import — so developers can run agents on sensitive data without exposing client records, credentials, or proprietary code to models, clouds, or MCP servers.

Three guarantees

1. The agent works fully — capability unchanged; scores, matches, charges, and aggregates complete normally.
2. The agent sees nothing — sensitive values stay encrypted; agents handle opaque tokens and references only.
3. The operator holds the keys — keys stay local, auditable, and revocable.

Install

export NXD_OPERATOR_PASSPHRASE="choose-a-long-random-passphrase"
pip install nxd

Requires Python 3.10 or 3.11 (Concrete ML FHE dependency).

Quick start

import nxd

# 1. Shield code before any AI call
code = "api_key = 'sk_live_xxxx'"
shielded = nxd.shield(code)
recovered = nxd.unshield(shielded)
print(f"AI sees:  {shielded[:40]}...")
print(f"You see:  {recovered}")
print(f"Match:    {code == recovered}")

# 2. Redact PII before sending to AI
note = "Patient John Smith, DOB 1985-04-12, SSN 432-11-5678"
clean, mapping = nxd.redact(note)
print(f"\nAI gets:  {clean}")
print(f"Original: {nxd.deredact(clean, mapping)}")

# 3. Vault a credential — agent never sees it
vault = nxd.Vault(agent_id="my-agent")
vault.store("stripe_key", "sk_live_xxxx")
result = vault.use("stripe_key", lambda k: f"charged via {k[:8]}...")
print(f"\nAgent got: {result}")
print("Key seen:  never")

# 4. Verify tamper-proof audit chain
nxd.audit.log("session", agent_id="my-agent")
print(f"\nAudit valid: {nxd.audit.verify()}")

Vault safety

⚠️ IMPORTANT: vault.use() protects your credential in storage and transit. It cannot prevent your callback from returning plaintext.

Safe: vault.use("stripe_key", lambda k: stripe.charge(k))

Unsafe — leaks key to caller: vault.use("stripe_key", lambda k: {"key": k, "status": "ok"})

Rule: callbacks should USE the credential, not RETURN it.

Redaction

redact() returns both the safe text and the local restoration mapping:

safe, mapping = nxd.redact("Patient John Smith, SSN 432-11-5678")
restored = nxd.deredact(safe, mapping)

mapping stores the original values locally so you can restore them with nxd.deredact(safe, mapping).

Handoff tokens

Handoff tokens are single-use. A token that has already been unpacked raises ReplayError if it is unpacked again.

handoff = nxd.Handoff()
token = handoff.pack({"client": "Jane Doe", "balance": 50000})
payload = handoff.unpack(token)

For multi-agent workflows where the same context is needed by multiple agents, pack a separate token for each agent.

Audit export

# Export a compliance report
nxd.audit.export("audit_report.json")
# Creates audit_report.json with full chain proof
# No sensitive values included in export

Benchmarks (MacBook Air, Python 3.11, Concrete ML 1.9.0)

Operation	Latency	Notes
FHE score (1 record)	~183 ms	First-call cold start
FHE score (1k records, parallel)	1.6 s	8 cores, ~1.6 ms/record
FHE match (single pair)	352 ms	Cross-system comparison
FHE aggregate (1k records, parallel)	1.8 s	~0.009% quantization error
Credential vault use	<1 ms	Decrypt in memory only
Proof suite	85/85 passed	python3 prove.py

What NXD does not protect against

NXD protects credentials and sensitive data from AI providers, model context, and ordinary cloud exposure. It does not remove the need for normal endpoint security and key management discipline.

If your local machine is compromised, master.key can be stolen. NXD protects credentials in transit and keeps them out of agent-visible plaintext, but it does not protect against local machine compromise.

NXD can prevent a model from seeing plaintext inputs. It does not control what a model does with the encrypted or redacted results it receives, so output handling still matters.

NXD uses FHE for specific compute operations such as score, match, and aggregate. It does not run the full LLM context window under FHE. For prompt and code protection, use redact() and shield().

The local master.key model is suitable for development and small deployments. Production systems should use a managed key system such as HashiCorp Vault or AWS KMS. Hosted key management is on the NXD roadmap.

NXD helps protect against external providers and cloud exposure. It does not protect against a trusted operator with physical access, because that operator holds the keys by design.

Current encryption choices are not presented as quantum-resistant. Post-quantum primitives are on the roadmap, but they are not part of the current release.

redact() is best-effort pattern detection for common PII and secret formats. It reduces exposure, but it is not a guarantee that every sensitive value in every format will be detected.

blur() uses calibrated Laplace privacy noise with explicit epsilon and sensitivity inputs. The helper is suitable for internal privacy-noise workflows, but formal differential-privacy claims for regulated deployments should follow an external review of the implementation and your parameter choices.

split() adds tamper detection to Shamir-style key splitting, but formal secret-sharing assurances should likewise be covered by external review before carrying a security certification claim.

NXD has completed an internal security review including:

• Adversarial testing of all primitives
• Cross-validation of Shamir implementation against sslib
• Statistical validation of differential privacy (KS-test vs diffprivlib)
• Randomness audit (no weak RNG found)
• MCP plaintext regression confirmed closed

split and blur are pending external cryptographic review. All other primitives wrap vetted libraries (Fernet, Ed25519, PBKDF2, Concrete ML).

Operator workflow

Set NXD_OPERATOR_PASSPHRASE before using the vault, audit chain, signatures, or any operator-only reveal flow. NXD stores ciphertext at rest, and the local key files are now wrapped with a PBKDF2-derived key from that operator passphrase.

When you use nxd init, NXD can vault .env secrets, replace them with NXD_VAULT::NAME references, and write an encrypted .env.backup.nxd recovery file.

On the MCP path, decrypt-style tools such as nxd_unshield, nxd_unseal_text, and nxd_detokenize no longer return plaintext to the agent. They queue an operator-only reveal:

nxd reveal <reveal_id>

Development

git clone https://github.com/Nexploraai/nxd
cd nxd
pip install -e ".[dev]"
python3 prove.py
python3 agent.py
python3 demo.py

License

Proprietary — Nexplora Labs. Free to use in projects, but the source may not be modified, redistributed, resold, or used to build a competing encryption or agent-protection product. See LICENSE.