SEDS · BUILT BY NAGE

Every decision sealed.
Every claim attributed.

Provenance-first AI workspace for legal and finance teams. seds turns every response into an attested record — STEMMA attribution, cryptographic seal, and a full audit trail. Where generic chatbots fabricate citations, seds extracts mechanical evidence from the model's own attention heads.

Q3 2026 launch · pilot pricing for design partners
WHY SEDS

Sealed receipt

Every response is delivered with an SHA-256-signed m5.9 receipt. Tampering is byte-detectable. Compatible with EU AI Act and SOC 2 audit-log requirements.

Mechanical attribution

STEMMA distributions are read from the model's own attention heads — not heuristic citation, but mechanical provenance.

Built-in counter-evidence

The model structurally surfaces counter-evidence beneath each claim. AI that knows when to say "we don't know."

INTERFACE PREVIEW
COMPARE · 14:28 · 312 tok · 1840 ms
94 AGROUNDED◆ b3f2

Compare CLAST against alternative initialization approaches

CLAST
RANDOM
LORA
Convergence
120× faster
1× baseline
~2× partial
VARVE isolation
Full
None
Partial (adapter)
Attribution
STEMMA — per query
None
None
TOKEN ATTRIBUTION · 63 STEP
Live demo →Mock data — production cutover once nage SSO is wired
OPEN STANDARD · KLM v1.0

seds doesn't just produce PDFs. It produces a .klm.

Knowledge Lineage Manifest — an open, portable, self-contained container for every AI-generated report. Where a PDF carries only content, a KLM also carries provenance: per-block receipts, STEMMA distributions, gamma vectors, VARVE signatures, the template snapshot, and an Ed25519 seal.

Standard PDF
  • Content
  • STEMMA distribution
  • Per-block receipt SHA
  • Template lineage
  • Offline tamper detectionpartial
  • Web citation archive
.klm bundle
  • Content (PDF + HTML)
  • STEMMA distribution
  • Per-block receipt SHA
  • Template lineage
  • Byte-precise tampering
  • Pin-on-export web archive
VERIFY FROM FOUR SURFACES
Web verifierseds.ai/verify
CLIpip install seds-klm
Python SDKfrom seds_klm import KLM
VS Code.klm opens inline
Verify a .klm file →KLM v1.0 spec ↗Apache-2.0 · IANA registration in progress: application/vnd.seds.klm+zip
VERTICAL PACKS

Want to pilot seds with your team?

We are running early pilots with law firms, financial advisors, and insurance carriers. Three months at no cost, unlimited queries, the Legal and Finance Packs, and a custom VARVE trained on your corpus. 50% off founding-customer pricing in exchange for structured feedback.