assay-ai 1.4.0

Receipt-native AI safety toolkit

Assay

Tamper-evident audit trails for AI systems. Logs record what you say happened. Assay makes the record tamper-evident, completeness-checkable, and independently verifiable -- including by someone who does not trust you. Integrity PASS + claims FAIL is an honest failure: authentic evidence that controls were violated. Assay does not prove external truth; it proves evidence integrity and makes omission detectable. Easier than a spreadsheet, harder to bullshit.

pip install assay-ai && assay quickstart

60-Second Demo

No API key needed. This runs a two-act scenario with synthetic data:

• Act 1: Agent uses gpt-4 with a guardian check. Result: integrity PASS, claims PASS.
• Act 2: Someone swaps the model and drops the guardian. Result: integrity PASS, claims FAIL.

That second result is an honest failure -- authentic evidence proving the run violated its declared standards. Not a cover-up. Not theater. Exit code 1.

assay demo-incident     # two-act scenario: honest PASS vs honest FAIL
assay demo-challenge    # optional CTF: spot the tampered pack
assay demo-pack         # optional: build + verify from scratch

The Golden Path

# 0. See Assay in action (recommended first step)
assay quickstart
# Or: assay onboard .  (guided project setup with doctor + CI guidance)

# 1. Find uninstrumented LLM calls
assay scan . --report   # generates a self-contained HTML gap report

# 2. Instrument (one line per SDK, or auto-patch)
assay patch .
#    from assay.integrations.openai import patch; patch()

# 3. Produce a signed proof pack
assay run -c receipt_completeness -- python my_app.py
# Add -c guardian_enforcement when you have a policy gate

# 4. Verify + explain
assay verify-pack ./proof_pack_*/
assay explain ./proof_pack_*/

# 5. Lock the verification contract
assay lock write --cards receipt_completeness -o assay.lock
assay verify-pack ./proof_pack_*/ --lock assay.lock --require-claim-pass

CI Gate (three commands, three exit codes)

assay run -c receipt_completeness -- python my_app.py
assay verify-pack ./proof_pack_*/ --lock assay.lock
assay diff ./baseline_pack/ ./proof_pack_*/ --gate-cost-pct 25 --gate-errors 0 --gate-strict

The lockfile catches config drift. Verify-pack catches tampering. Diff catches regressions and budget overruns. See Decision Escrow for the protocol model behind this workflow.

How It Works

Assay separates two questions on purpose:

• Integrity: "Were these bytes tampered with after creation?" (signatures, hashes, required files)
• Claims: "Does this evidence satisfy our declared governance checks?" (receipt types, counts, field values)

Integrity	Claims	Exit Code	Meaning
PASS	PASS	0	Evidence checks out, behavior meets standards
PASS	FAIL	1	Honest failure: authentic evidence of standards violation
FAIL	--	2	Evidence has been tampered with
--	--	3	Bad input (invalid arguments, missing files)

The split is the point. Systems that can prove they failed honestly are more trustworthy than systems that always claim to pass.

Commands

Command	Purpose
assay quickstart	One command: demo + scan + next steps
assay demo-incident	Two-act scenario: passing run vs failing run
assay demo-challenge	CTF-style good + tampered pack pair
assay demo-pack	Generate demo packs (no config needed)
assay onboard	Guided setup: doctor -> scan -> first run plan
assay scan	Find uninstrumented LLM call sites (--report for HTML)
assay patch	Auto-insert SDK integration patches into your entrypoint
assay run	Wrap command, collect receipts, build signed pack
assay verify-pack	Verify a Proof Pack (integrity + claims)
assay explain	Plain-English summary of a proof pack
assay ci init github	Generate a GitHub Actions workflow
assay lock write	Freeze verification contract to lockfile
assay lock check	Validate lockfile against current card definitions
assay cards list	List built-in run cards and their claims
assay cards show	Show card details, claims, and parameters
assay diff	Compare two packs: claims, cost, latency, model churn (--gate-* for CI thresholds)
assay analyze	Receipt-level time-series analysis
assay doctor	Preflight check: is Assay ready here?

Documentation

• Quickstart -- install, golden path, command reference
• Decision Escrow -- protocol model: agent actions don't settle until verified
• For Compliance Teams -- what auditors see, evidence artifacts, framework alignment
• Repo Map -- what lives where across the Assay ecosystem
• Pilot Program -- early adopter program details

Scan Study

We scanned 30 popular open-source AI projects for tamper-evident audit trails. Found 202 high-confidence LLM SDK call sites across 21 projects. Zero had evidence emission at any call site. Full results.

Get Involved

• Try it: pip install assay-ai && assay quickstart
• Questions / feedback: GitHub Discussions
• Bug reports: Issues
• Pilot integration: Want help instrumenting your AI stack? Open a pilot inquiry

Related Repos

Repo	Purpose
assay	Core CLI, SDK, conformance corpus (this repo)
assay-verify-action	GitHub Action for CI verification
assay-ledger	Public transparency ledger

License

Apache-2.0