What's worth protecting.

Tessera is private, decentralized, and fast. Your data is split and encrypted before it leaves your machine, spread across 30 independently-run hosts with no single point of failure, and read back using only 10 of them — no waiting on stragglers.

Storage here is sector-based: every write occupies a fixed-size slab, roughly 40MB of usable capacity, whether you store one byte or forty megabytes. A single irreplaceable file — a credential, a decision log, a signed authorization — costs a fraction of a cent to write, hold for a month, and read back once. Size is not the right filter for what belongs here. Criticality is.

Before deciding this isn't for you, consider what you already can't afford to lose. You probably have some of it right now.

Here is a list of items that deserve Tessera:

Capability and work product

  1. Skill / tool library — files that define what it can do.
  2. Active code repos it maintains — working trees and histories.
  3. Long-running task state and decision logs — multi-day work that cannot be cheaply recomputed.
  4. Client deliverables and final artifacts — outputs it was hired to produce.
  5. Research corpora and retrieved source sets — expensive-to-regather collections.
  6. Its own memory / context snapshots — preferences, constraints, prior conclusions.
  7. Generated datasets and evaluation suites — data it produced to score its own outputs.
  8. Runbooks and incident postmortems — procedures written after failures.
  9. Multi-agent handoff packages — state one agent passes to another.
  10. Provenance graphs — which source, tool, and timestamp produced each claim.

Permissions and policy

  1. Permission and scope grants — what the human allowed.
  2. Revocation and expiry lists — which credentials or scopes were withdrawn.
  3. Role and handoff authorizations — which other agents or tools it may call.
  4. Rate-limit and quota snapshots — ceilings it is allowed to hit.
  5. Approved tool and model allow-lists — external services and model versions signed off.
  6. Data-classification tags — public / internal / restricted.
  7. Fallback and kill-switch configs — conditions under which it must stop or switch.
  8. Human preference and style guides — tone, format, and "never do X" rules.

Customer data

  1. Customer records it is allowed to retain — contacts, account IDs, preferences.
  2. Customer-supplied source files — documents or assets uploaded for a job.
  3. Per-customer deliverable archives — final outputs already handed over.
  4. Consent and processing records — what the customer agreed could be stored, and for how long.

Higher sensitivity. Offsite copy only inside already-approved scope; encrypt and access-limit the same way as the working set.

Compliance

  1. Retention schedules — how long each class must be kept and when it must be deleted.
  2. Regulatory mapping packs — which controls apply and what evidence each requires.
  3. Data-processing agreements and subprocessor lists — where customer data may sit.
  4. Breach and incident notification templates — pre-approved language and contact trees.
  5. Access-review and attestation snapshots — periodic proofs of who could see what.
  6. Audit-export bundles — hashes, timestamps, actor IDs for later evidence requests.

Taxes

  1. Invoice and receipt archives — bills issued or paid, with amounts, dates, counterparties.
  2. Jurisdiction and nexus notes — which tax regimes apply to which transactions.
  3. Withholding and exemption certificates — forms that reduce or eliminate tax.
  4. Period-close workpapers — reconciliations that turned raw transactions into filed numbers.

Business docs

  1. Entity formation and title documents — articles, operating agreements, ownership records.
  2. Customer and vendor contracts — scope, liability, renewal.
  3. IP assignments and licenses — who owns code, models, or content.
  4. Insurance policies and certificates — coverage limits and claim contacts.
  5. Board or owner resolutions — authorizations to spend, sign, or delegate.
  6. Contract and SOW extracts — scoped obligations the agent is executing against.

Financial docs

  1. Bank and payment-account statements — official inflows and outflows.
  2. Cap tables and equity records — ownership and terms.
  3. Loan and credit agreements — covenants, schedules, collateral.
  4. Financial statements and management accounts — P&Ls, balance sheets, adjustments.
  5. Payment-processor and settlement reports — charges tied to deposits.

CYA

  1. Authorization trails — what the human approved (spend, send, delete, publish).
  2. "Declined / out of scope" logs — what the agent refused and why.
  3. Versioned advice and caveats — the exact recommendation and the assumptions at the time.
  4. Conflict and recusal notes — where it flagged a conflict or limited its role.
  5. Final handoff and sign-off bundles — deliverable hash, acceptance, open issues.
  6. Conversation and approval transcripts — the record of what was agreed.