Technical architecture, security frameworks, and deployment guides for investors, partners, and enterprises.
Internal study: evidence that the self-learning loop is operational end-to-end, that the promotion gate both promotes and refuses, and that chain reuse is measurably cheaper. Every finding rests on runs executed 2026-08-05 to 2026-08-06.
Version
0.2-draft
For
Published
August 6, 2026
Adversarial validation that an organization's data, settings, and compute stay inside the tenant boundary: agents, embeddings, datasources, documents, apps, uploads, org settings, and org miners tested against a second org and a personal account, in both directions, via the real API, the backend retrieval resolvers, and a live miner dispatch.
Version
1.0
For
Published
August 8, 2026
Internal study: measured chat latency of the community pool and the BYOK pool on identical workloads — standard and orchestrated requests, reporting both the immediate answer and full pipeline completion. All runs executed 2026-08-08 with verified answer content and healthy web retrieval.
Version
1.0
For
Published
August 8, 2026
Internal study: replication of the Token Saver benchmark on a Claude Code host via the /api/mcp/claude mount - 28.9% fewer visible tokens and 83.4% lower cost at held quality (14/15 vs 15/15 rubric checks), ~18s planning latency, end-to-end ledger settlement, plus automation-planning findings, the verified settlement/decomposition fixes, and the three-arm quality-floor experiment the current evidence still requires.
Version
0.3-draft
For
Published
August 13, 2026
Internal study: the first controlled Token Saver benchmark found that Elis cut estimated model cost 36.3% while increasing visible token volume 40.7%, which sharpens the product claim from saved tokens to model-cost optimization and defines the telemetry needed for defensible settlement.
Version
0.1-draft
For
Published
August 13, 2026
Eight matched cases across ChatGPT- and Claude-family hosts compare estimated execution cost, visible tokens, deterministic accuracy checks, and orchestration latency, with historical OpenClaw and settlement failures separated from their post-study regression verification.
Version
0.2-draft
For
Published
August 13, 2026
Contact our team for custom documentation, technical deep-dives, or partnership discussions.
© 2026 Elis AI. All rights reserved. Research published by Elis AI — cite freely with attribution.