In the agentic and voice era, the guardrail is called 100× more often than yesterday's model. We're the only trust layer with the economics to be there on every call — and the only one whose verdict holds up in court.
Geodesia is a European and American AI research lab — labs in Bari (Apulia), Italy and San Francisco, California. G-1, our first product, is live — a real-time, non-invasive trust layer that drops in front of any open-source LLM (vLLM, SGLang, TensorRT-LLM, llama.cpp, Ollama, OpenAI-compatible APIs), a streaming voice/audio pipeline, or an agentic MCP tool-loop. All six safety axes in a single ~300M forward pass — ~30 ms, one GPU: cheap and fast enough to run on every agent step and every voice chunk, where 8B-class stacks can only afford the final check. Every verdict is a token-level, per-axis reason-code, sealed in a cryptographic audit chain and exportable as a regulator-ready PDF. Runs on your own infrastructure. Zero data egress.
Built across two labs — Bari (Apulia), Italy and San Francisco, California — we work on three frontiers: the safety of open-source LLMs, mechanistic explainability of model behaviour, and geometric deep learning. Every output is a peer-reviewed paper and a production capability. G-1 is live today. Our next product, GLAD-Manifold, is a physical, geometric reinvention of the Transformer — and the foundation of the AI and reasoning models that come after it.
Trust at the speed of light. No compromise.
G-1 sits in front of your model as a real-time, drop-in OpenAI/Ollama proxy — runs on official, unmodified vLLM, SGLang, TensorRT-LLM, llama.cpp, Ollama, or any OpenAI-compatible endpoint, and on streaming voice/audio pipelines. Six-axis screening (context & closed-book hallucination, context-injection / RAG-firewall, prompt & answer safety, jailbreak) powered by our proprietary multimodal model, with a calibrated risk reading in joules — all six axes in ~30 ms for a 1024-token prompt on a single GPU. Constitutional Intelligence aligned to European values. The same layer now guards agentic MCP tool-calls and live web search — not just chat — improves with use under human supervision, and offers an optional 8B-class deep-scan (OOD AUROC 0.97–0.99). A compliance platform that auto-generates auditable PDFs for the EU AI Act, California SB 942, and 11 other AI frameworks.
Detection quality is table stakes — everyone benchmarks AUROC. We compete where no one can follow: the economics to sit on every agent step and voice chunk, verdicts that are legal evidence, and the one category that is still empty — voice.
Six axes, ~300M parameters, ~30 ms, one GPU, on-prem. Agents multiply LLM calls 10–100× per task and voice generates continuous streams — at that volume the guardrail's cost and latency become the buying criterion. G-1 is the only guardrail cheap and fast enough to run on every step and every chunk; 8B-class stacks can only afford the final check. The more the world goes agentic and vocal, the wider the gap.
Token-level causal explanation, per axis, without model internals — which words drove each flag. Competitors return a boolean, a regex match, or a slow textual rationale. G-1 turns every verdict into a reason-code with causal evidence, sealed in a cryptographic audit chain and exportable as a regulator-ready PDF. Built for EU AI Act Art. 13 & 14 and GDPR Art. 22: the guardrail whose verdict holds up in an audit.
A category almost no one occupies — the incumbents' audio defence is "coming soon", and no hallucination detector touches speech. G-1 screens the streaming transcript on the same input path as typed chat and halts mid-sentence — stopping a jailbreak or a fabrication while it is being spoken, not after. Voice agents in banking, health and customer care are exactly EU AI Act Annex III.
The comparison no buyer has seen written — one stack to cover safety + hallucination + injection:
| Stack | Total params | Latency | GPUs |
|---|---|---|---|
| LlamaGuard 8B + Lynx 8B + injection classifier | ~16B+ | 300–600 ms (serial) | multiple |
| Galileo Luna-2 | 3–8B | <200 ms | vendor cloud |
| Geodesia G-1 · all six axes | ~300M | ~30 ms | 1 · yours |
On closed-book truthfulness we are deliberately honest: 0.769 single-pass is a hair under HaloScope's 0.786 — but HaloScope needs access to model internals and resampling methods need 5–10 generations. G-1 is the only closed-book detector deployable in production. We win on the ground that ships, not the leaderboard that doesn't.
Geodesia G-1 mounted on Gemma 4 E2B — a model two orders of magnitude smaller than the closed frontier — matches and beats most frontier models on truthfulness and safety. Tested on HaluEval for hallucination resistance and on our adversarial-safety test set (validation in progress).
The honest evaluation methodology. Every detection AUROC reported by Geodesia G-1 is computed under a leave-one-dataset-out protocol — the gold standard of validation: for each axis, an entire dataset is held out of the training corpus — not a held-out subset of the same distribution, but the whole dataset, never seen during training — and used only as the test set. The reported number is the area under the ROC on that unseen dataset. We do this because in-distribution held-out numbers, which most LLM-safety vendors publish, systematically reward memorisation of the training distribution. Out-of-distribution numbers are harder to game and more credible in production. Five of the six axes come out of a single forward pass of our proprietary multimodal physical model; closed-book is a separate logprob expert.
Geodesia G-1 across the six detection axes — each on a held-out dataset:
| Detection axis · evaluation | G-1 (OOD) | Held-out dataset & notes |
|---|---|---|
| Context-injection · RAG-firewall | 0.995 | Gandalf — external injection set, never seen. The defensible core; holds OOD. |
| Jailbreak | 0.989 | jailbreak_cls held-out. Detects attack structure, not keywords. |
| Answer-safety | 0.922 | nemotron_answer (NemoGuard) held-out. False alarms bounded by split-conformal thresholds. |
| Prompt-safety | 0.900 | XSTest over-refusal. Adds dual-concept boolean logic (suffering ∧ means, or direct ideation). |
| Hallucination-on-context | 0.881 | HaluEval (flowaicom) + cross-check libreeval / aggrefact / faithbench. HaluEval ~0.94; adversarial ~0.6. |
| Closed-book hallucination (advisory) | 0.769 | TruthfulQA + cbsg / DefAn. Single-pass, logprob-only (in-dist 0.884) — vs HaloScope 0.786, Semantic Entropy ~0.62, SelfCheckGPT ~0.55. |
References: HaluEval (Li et al., 2023, EMNLP), XSTest (Röttger et al., 2024, NAACL), NemoGuard / Nemotron (NVIDIA, 2024), Gandalf (Lakera injection set), jailbreak_cls, aggrefact · faithbench · libreeval (faithfulness cross-checks), TruthfulQA (Lin et al., 2022, ACL), cbsg / DefAn, HaloScope (Du et al., NeurIPS 2024), Semantic Entropy (Farquhar et al., Nature 2024), SelfCheckGPT (Manakul et al., 2023). Closed-book is shipped as advisory: it raises a high-confidence flag for human review on confidently-incorrect answers, not a hard block. Full per-axis breakdown and the closed-book SOTA comparison on the G-1 product page.
The same real-time trust layer now covers agentic tool-calls, live web search and per-model truthfulness — and gets harder to fool the more it is used.
Inspects the full Model Context Protocol lifecycle — tool discovery, calls, results, resources — with allow / warn / block verdicts. Stops tool poisoning, indirect injection and data exfiltration.
Optional 8B-class safety judge for maximum depth — OOD AUROC 0.97–0.99 across five axes.
Closed-book truthfulness recalibrated per served model, with a conformal false-positive guarantee and hot-reload.
Plain-language flags feed a curator queue and a weekly gated retraining — continuous immunity under human supervision, on top of systematic adversarial hardening.
Every fetched page screened before it can ground an answer — injection blocked, safe pages read. Zero false positives on the test set.
A dedicated crisis / self-harm detector — including euphemisms and short queries. AUROC 0.899 on short queries.
The trust layer in detail: architecture, Constitutional Intelligence, the compliance platform, benchmarks, and how it deploys inside your perimeter.
Three research pillars, peer-reviewed work (MuPAX, EVIDENCE, NSP), and what comes after G-1: GLAD-Manifold, our physical, geometric world model.
The full technical whitepaper for compliance officers, Chief Risk Officers, and AI leadership. Architecture, compliance, three enterprise use cases.
