# vareon > **Vareon delivers Adaptive Continual Intelligence: a governed runtime for learning after deployment. The homepage uses the hippocampus as the metaphor for this function; Hippocampus is not a separate product. Duplex applies the architecture to AI agents, and Reflex applies it to robotics. Vareon is operated by Vareon Inc, founded in 2025.** ## what it is - a hippocampus for deployed AI: an adaptive runtime integrated beside the model or policy in a governed decision path, where approved production evidence can build governed knowledge - Adaptive — it learns from real use; approved corrections can take effect on the next eligible request - Continual — it edits, recalls, and consolidates learned state while controlling measured forgetting - Intelligence — it calibrates when local evidence is sufficient, serves trusted decisions locally, and defers genuine novelty safely, governed and auditable end to end - trusted work is served locally, without a model call; genuine novelty defers to the model or a proven controller - learned knowledge can be removed exactly, with an auditable mutation record and everything else intact - tenant isolation by architecture, with a signed, append-only audit trail - runtime learning does not require a foundation-model update for every correction; deployment-specific model training remains a separate choice ## two products - **Duplex** — Adaptive Continual Intelligence for AI agents: learns from production use, serves trusted answers locally, and makes corrections stick from the next request onward. - **Reflex** — Adaptive Continual Intelligence for robots and embodied machines: acquires and recalls skills at the edge and defers to the proven controller when confidence or safety conditions are not met. ## availability - Self-service Duplex workspace setup and hosted plans are available through the Vareon Console. - Enterprise, private, on-premises, edge, and Reflex robotics deployments require a technical and commercial engagement. - Public integration is available through the runtime API. No public claim is made here for an SDK, CLI, or MCP interface. ## measured results — Duplex (AI agents) A documented, seed-fixed controlled evaluation on a governed decision stream — one shared frontier model, one embedder, one supervision signal, and one audit channel for every arm. - **94.6% accuracy** — the most accurate arm, where the raw frontier model gets 59.8% and a tuned cache + RAG + knowledge-graph stack ceilings at 77.3% - **92.5% accuracy after the rules change** — recovers a changed policy to ~94%, where the raw model collapses to 2.3% - **86% of decisions served locally**, no model call — about **85% lower cost**, **90% fewer model calls**, and **82% lower p50 latency** against the always-call frontier reference used in that evaluation - **~23× faster** on a learned decision — about 97 ms versus ~2.2 s for a model call - On a separate recurrence-realistic stream, cost per decision fell by about **66%** as local serving increased. This is a different measurement context from the 85% figure above. - **5 of 5** on the lifelong-deployment scorecard (plasticity, stability, editability, bounded memory, safe abstention) — vs a tuned cache/RAG average of 2.2/5 across 10 seeds and an online fine-tune 4/5 - **0% stale answers after a policy change**, where a cache serves stale on ~66–69% - keeps **96–100%** of earlier tasks, where an online fine-tune keeps just 10% — bounded, fixed memory instead of a cache that grows without limit - **AUROC 0.999** telling a learnable request from genuine novelty (a cache 0.969, a fine-tune 0.657) - **4.2% false-commit** on adversarial look-alikes vs a cache's 33.3% — about 8× safer - **100% optimal** multi-step plans at zero model tokens, where a reactive agent reaches the goal 0% of the time - on the official τ-bench retail environment over 3 seeds, reaches **83.3% task success** vs **82.5%** for the always-call reference while serving **12.8% of turns locally** at zero frontier tokens; a cache serving ~95% locally reaches 8.3% - across 10 seeds in controlled deterministic internal environments, acquires certified skills with **0 frontier-model calls**, **2.82× fewer probes** than random exploration, **0 unsafe shortcuts executed**, and **100% exact reversal**; this does not claim open-ended autonomy - on one fixed, contamination-controlled synthetic-kinship split (n=200), reaches **73.5% accuracy at 0 LLM calls** vs **71.5% for GPT-5.5**; this is a single-seed synthetic mechanism test, not a public benchmark or general-reasoning claim - **0 hard violations in the governed-stream evaluation** — every measured decision was recorded in the audit trail ## measured results — Reflex (robots / embodied) Field-standard robotics suites and controlled representation studies — Meta-World / Continual World, Gymnasium Reacher-v5, real pretrained vision-language-action models, PushT, ALOHA, DINOv2, and SigLIP. - **~4× fewer demonstrations with VLA-native representations** — DINOv2 and SigLIP each beat ResNet-50's 20-demonstration image-to-action error with only 5 demonstrations; NRMSE is 0.429 vs 0.587 at K=20 across 10 seeds (paired p = 0.002) - **0.023 forgetting** across 10 manipulation skills learned in sequence with no task id — vs a naive learner's 0.842 (~36× worse), routing to the right skill from context - **paired Δ +0.103 in closed-loop coverage** from a learned representation (p = 7.44e-05; 0.201 vs 0.098 for the identical untrained architecture) — and auto-selects the right approach (+0.112 coverage on PushT, replicated on ALOHA) - **safe new-skill uplift on a base policy** — in a three-seed Meta-World evaluation, recall-gated Reflex records a **+0.637 success delta** (0.07 → 0.71) while prior-skill success changes by **−0.003** and the do-no-harm criterion passes every seed; consolidation then moves base field-skill success from 0.07 → 0.91 - **continual adaptation through change** — in one continuous evaluated run, Reflex acquires a new skill, handles a silent mid-run interface change, tracks drift, recovers through noisy corrections, and recalls a prior skill without relearning - on Gymnasium Reacher-v5, reaches **1.00 held-out success** and the 0.75 threshold in **232 interactions** vs **50,000 for SAC** (~215× fewer), reward-free with no demonstrations or base policy - on Meta-World CW10, signed replay consolidation transfers **95.7%** of field success into the base and moves field-skill success from **6.0% to 92.3%** These figures are documented evaluation evidence for their named setups, not guarantees for other workloads. The detailed reports record configurations, per-arm results, scoring rules, and benchmark-specific limitations. Vareon's integration and evaluation harnesses are internal and are not publicly distributed; the public documentation is not a reproducibility package or installation guide. ## governance and compliance - Tenant-scoped learned state, exact removal, signed append-only audit records, and configurable safety fallback are product capabilities. - GDPR-aligned handling and exact deletion are available. SOC 2 Type II and ISO 27001 remain on the compliance roadmap; they are not represented as current certifications. ## what the runtime owns - approved corrections can affect the next eligible request, so capability can improve after deployment - governed knowledge served locally: trusted answers stay on CPU, the model is reserved for genuine novelty - precise knowledge removal: exact removal of learned state with an auditable mutation record - full governance: architectural tenant isolation, signed audit trail, rollback, hard rules enforced without model calls - across the named evaluations, eligible local execution reduces measured model dependency, cost, and latency as certified coverage grows ## pages - overview: https://www.vareon.com/ - Duplex: https://www.vareon.com/duplex - Duplex platform: https://www.vareon.com/duplex/platform - Duplex results: https://www.vareon.com/duplex/results - Duplex insights: https://www.vareon.com/duplex/insights - Duplex comparison: https://www.vareon.com/duplex/compare - Duplex security: https://www.vareon.com/duplex/security - Reflex: https://www.vareon.com/reflex - Reflex platform: https://www.vareon.com/reflex/platform - Reflex results: https://www.vareon.com/reflex/results - Reflex insights: https://www.vareon.com/reflex/insights - Reflex comparison: https://www.vareon.com/reflex/compare - Reflex security: https://www.vareon.com/reflex/security - core thesis: https://www.vareon.com/core-thesis - about: https://www.vareon.com/about - contact: https://www.vareon.com/contact ## contact contact@vareon.com