Laboratory instruments
Small surfaces. Explicit boundaries.
YY builds focused products and open instruments for examining datasets, records, and AI-system behavior. Across the portfolio: explicit data boundaries, reviewable evidence where claims matter, and a visible limit on every result.
- Keep data close
- Show the evidence
- Name the limit
The release pattern
The omissions are part of the design
- 01
Keep data close
File-inspection paths run on your machine or in browser memory. Platform utilities state what the host presents and what YY receives.
- 02
Show the evidence
Samples, schemas, manifests, or source leave a smaller result another reviewer can inspect.
- 03
Refuse broad verdicts
A hash, receipt, or policy pass stays a bounded observation—not a safety, compliance, or correctness certificate.
Paid tools
Buy the narrow result, not a promise cloud
One-time founding licenses plus one metered Poe utility. Inspect the public sample, measured scope, price, and non-claims before paying.
50% off AgentSafe and Dataset Preflight through .
- Paid · local
AgentSafe
A deterministic, local evidence checker for one specific behavior: approval-to-execution in OpenAI Agents SDK 0.19.x FunctionTools.
One measured approval-to-execution behavior—not a security audit or certification.
- $49 → $24.50 through Aug 18
- 0.0.1-rc7
- SDK 0.19.x
- Paid · local
Dataset Preflight
Inspect a local dataset for duplicates, provenance gaps, and secret-risk signals, enforce your thresholds, and emit opt-in SARIF and JUnit evidence for CI.
A bounded evidence packet—not a safety, license, or training-readiness verdict.
- $39 → $19.50 through Aug 18
- v0.2.0
- SARIF + JUnit
- Paid per message · Poe
Chat Receipt
Export the conversation visible to the bot as a deterministic ZIP with a readable transcript, canonical JSON receipt, SHA-256 manifest, and local verifier.
Exports only the history Poe presents to the bot; the unsigned receipt does not prove truth, identity, origin, or completeness.
- 500 YY points + Poe base
- No model call
- 20 tests
Free and open work
Start with something inspectable
Inspect a research model, review a specification through two named models, use the browser verifier, or install a release.
- Open research · Python + RTL
Analog Matmul
A behavioral mixed-signal matrix model, held-out 4 × 4 characterization, driver contract, and open-silicon roadmap for inference and hybrid-training research.
Behavioral model and interface contract only—no transistor design, layout, fabricated silicon, or measured performance.
- CERN-OHL-P-2.0 + Apache-2.0
- v0.2.0
- 48 tests
- No YY fee · Poe
Spec Court
Send one bounded software or product specification to two named reviewer models and download their independent findings as a deterministic, locally verifiable receipt.
Preserves two model reviews and their limits; it does not certify that either review is true, complete, secure, or correct.
- Two named models
- Digest consent
- 20 tests
- Free · browser memory
Evidence Verifier
Recompute every SHA-256 digest in a supported evidence packet and inspect the result without uploading the files.
Sends YY no file names, contents, hashes, or verification results.
- Web Crypto
- No account
- No upload
- Open source · Node.js
ClipCheck
Offline, deterministic preflight checks for short-form video files before they are uploaded.
Checks declared media rules; it does not judge creative quality or platform performance.
- Apache-2.0
- v0.1.0
- Windows + Linux CI
- Open source · Python
Workload Receipt
Deterministic, content-free receipts for observed AI workload usage metadata, with explicit evidence gaps and later integrity verification.
Retains no prompts or tool payloads; a receipt is not a compliance certificate.
- Apache-2.0
- v0.1.0
- PyPI
- Open source · Python
Corpus Evidence
An open-source Python library that inventories a local corpus, incorporates optional provenance declarations, and produces a deterministic manifest that can be checked for drift later.
Records evidence and human declarations; it does not make legal conclusions. Alpha supports CPython 3.11 on Windows only.
- Apache-2.0
- v0.1.0
- PyPI
- Windows · CPython 3.11
- Open source · Zig
Chi
Streaming, checkpointable Zig components for bounded corpus ingestion, text cleanup, document filtering, and exact or approximate deduplication.
Pre-1.0 components with explicit ownership and versioned checkpoint formats.
- Apache-2.0
- v0.1.0
- Zig 0.16.0
YY IDE
An agent-first development environment where the agent operates the loop and the human supplies intent, judgment, approvals, and credentials.
Private V0 prototype. The capture below is the real React interface running deterministic fixture state—not a constructed mockup or a launch announcement.
See the working V0 interface