Transparency & Verification

This website is structured to match how people explore ideas. We separate the human experience from the engineering underneath:

• The Surface (plain white): The human experience in plain language. How algn feels to use, what we believe, and the choices we make. No jargon and no unexplained system names.
• The Depth (technical dark): The mechanics. How data travels, where encryption keys live, how background agents coordinate, and what each choice costs.
• Whitepapers & Research (/research): Academic papers with numbered claims, citations, formal threat models, and explicit statements of what the research does not prove.

Every content page and research paper carries a public review status at the top, linking directly to a review ledger and correction form at the bottom. Anyone can challenge our claims, submit missing sources, or flag errors.

This website reflects what actually shipped in the build today.

When code ships or policies change, public documentation updates to match the current build. Public review ledgers on each content page record changes and verified corrections over time, keeping the platform accountable to the people who use it. Research papers remain formal, static academic records with published versions and errata ledgers.

Pending release Open repository pending release.

Most deployed AI memory treats personal reflections as an append-only cloud database: flat lookup tables that hoard personal data on remote servers and degrade over time.

algn built a different foundation. Sera’s memory is a biological-inspired, local-first cognitive system that consolidates similar reflections, fades situational noise while preserving core values, clusters themes across weeks, and operates entirely on your device under keys only you hold.

Because solving longitudinal, privacy-preserving memory is a foundational challenge for human-centred computing, we are releasing the vault core as an open-source project.

• Verifiable privacy: Security researchers can inspect the local key generation, encryption primitives, and deletion routines to confirm that memory keys never leave the device.
• Public infrastructure: Universities, research institutions, grant-backed initiatives, and independent developers can reuse the memory architecture for their own use cases without needing to build centralised surveillance infrastructure.
• Data-flow manifest: Every background process, the data it touches, where inference occurs, and its zero-retention guarantee is published in an inspectable manifest.

Claims on algn are designed to be checked directly rather than taken on trust.

• Encrypted-at-rest inspection: Free text in the vault inspection view is displayed as a SHA-256 hash and byte length by default. Revealing a record displays the raw initialisation vector and record-bound AES-256-GCM ciphertext stored on your device alongside its plaintext hash.
• Outbound network receipts (Egress): A rolling 24-hour log records every request leaving your device, showing Measured client payload bytes and SHA-256 hashes (which you can match against your browser’s developer tools) side-by-side with server Declared zero-retention hop receipts.
• On-device embedding check (Clusters): You can inspect the 384-dimension vector for any memory and run a local Re-embed check to verify that cosine similarity matches (1.000) on-device without network traffic.
• Metacognitive and voice logs (Meta and Voice): A rolling 24-hour log displays background pattern analyses, safety grounding, and Be readings, including the exact note passed to Sera on each turn.
• Key integrity and crypto-shredding (Keys and Sync): You can verify that your browser’s Web Crypto keys are non-extractable, test your passphrase against its key check value, or inspect the .algn backup file. Deleting your vault permanently destroys the encryption key (crypto-shredding).
• In the code: The open-source memory vault repository lets cryptographers and developers audit the cryptographic implementation (AES-256-GCM via native Web Crypto) and verify that no telemetry or backdoors exist.
• In the research: Every research paper publishes its methodology, citations, and explicit limits. If our data or reasoning is flawed, the open review ledger records the correction publicly.

Unconditional disclosure is reckless. Total secrecy is extractive. algn draws an explicit boundary between what must be public and what must be protected.

Our standard is simple: we publish what the system does and why it works; we withhold what would help a bad actor break it or a competitor clone it.

Publishing safety thresholds would provide an instruction manual for malicious actors looking to bypass crisis protocols or harass matched members. Keeping those heuristics private protects the people inside the community. Similarly, proprietary prompt steering assemblies represent years of careful research and engineering that make Sera unique, and are kept to protect the organisation’s viability.