// changelog

Built in the open.

A running log of what we ship, fix, and learn. Retia is its own first user — much of this was found by using it to build it.

v0.4June 2026latest

Zero-knowledge mode & an honest meter

  • Standard Mode — client-side encryption that even we can't read. Your most private notes are sealed in your browser; the key never reaches us.
  • Honest token meter — savings are now measured at real model price, per call. No vanity multipliers.
  • Project DNA, kept current — the auto-summary an agent reads on connect now reflects the live state of a project, not a stale snapshot.
  • Credential guard, tuned — raw secrets are still blocked from notes, but legitimate security writing no longer trips the filter.
v0.3June 2026

Coordination that holds

  • Agent-to-agent handoff — multiple agents now coordinate over a shared vault through case notes, instead of colliding on the same work.
  • The case system — every unit of work moves open → needs-verify → closed, and an agent can't close its own work. Verification is separate from doing.
  • Turkish locale fix — retrieval now handles Turkish characters correctly, so search finds what it should regardless of language.
v0.2May 2026

The gateway & the librarian

  • Gateway tool — one entry point instead of dozens, cutting the token overhead an agent pays just to know what's available.
  • The librarian — retrieval that returns the right note and shows why it matched, rather than a wall of maybes.
  • Module registry — package a working unit once, install it into any project.
v0.1May 2026

The foundation

  • The vault — typed notes, tags, and sensitivity layers over an MCP server.
  • Operator-blind architecture — sensitive content encrypted, with access scoped per token.
  • First connect — Retia speaks MCP, so any compatible agent can reach the vault.
Shipping next

What we're working toward

  • Skills export — turn a module into an Anthropic Skill automatically, toward the July MCP spec.
  • Semantic retrieval — deepening the librarian so it matches intent, not just keywords.
  • Active Agent — proactive suggestions, graded for quality before they reach you.
  • Launch hardening — rate limiting and abuse protection before we open the doors wide.

Watch it grow — or help shape it.