RC / RootCompute / Pre-release combined offering
RootCompute brings inVtero hardware-first runtime introspection together with XAMMY watermarking and provenance so supported workloads, observed code and data, and generated information can be investigated in one evidence workflow.
RootCompute is a pre-release combined offering. Observation coverage depends on the acquisition path, supported representations and configured access. Provenance is evidence of origin or association, not a substitute for independent quality, truth or policy evaluation.
RootCompute / Code & Data Evidence Fabric
Evidence from compute to content.
RootCompute brings inVtero hardware-first runtime introspection together with XAMMY watermarking and provenance so supported workloads, observed code and data, and generated information can be investigated in one evidence workflow.
RootCompute is a pre-release combined offering. Observation coverage depends on the acquisition path, supported representations and configured access. Provenance is evidence of origin or association, not a substitute for independent quality, truth or policy evaluation.
01 / Runtime and data sensing
Reconstruct supported execution environments and inspect accessible code and data.
Physical-memory analysis, page-table reconstruction, executable-page verification and text/byte/word-pattern sensing form the observation foundation.
02 / Information provenance
Mark, evaluate and preserve evidence associated with generated information.
Generation-time watermarking, transformation evaluation and configurable signed provenance records provide information-side evidence.
Connected evidence workflow
01
Acquire supported workload evidence while keeping access and coverage boundaries explicit.
02
Recover address spaces, executable pages, data representations and workload context.
03
Connect available generation signals, provenance records and observation context without conflating them.
04
Use the resulting evidence for integrity, data-protection, investigation and independent reuse decisions.
Our portfolio
Runtime integrity and AI provenance form the core of our portfolio. Each addresses a different evidence problem; neither substitutes for the other.
Memory integrity & machine-state analysis
A composable integrity engine for bare-metal, virtual, mixed host/hypervisor, nested page-table and SmartNIC-connected topologies. Reconstruct memory context and validate executable code within the pipeline you choose.
Foundation: physical-memory analysis, recursive introspection, relocation-aware verification and programmable workflows.
AI direction: structured evidence, semantic change analysis and grounded investigation.
LLM watermarking & provenance
A research toolkit for watermarking and provenance in generated language. Bring generation, detection experiments and inspectable artifacts into one workflow.
Platform: vLLM on NVIDIA CUDA and MLX on Apple Silicon.
Status: research, pre-release. Evaluate detection and text quality against your own traffic and transformations.
From the lab
Explore the next product directions, with the current foundation and future plans kept distinct.
Local inference infrastructure
Alpha / in developmentYour local inference fleet.
An open-source foundation for bringing multiple Ollama instances behind one endpoint. A playful name for a practical infrastructure problem.
Next-release discussion: real cross-host layer handoff, capability-aware placement and predictable failure behavior. Forward-looking work, not shipping guarantees.
Agent workflow visibility
Coming soonA closer view of agent-assisted work.
A forthcoming toolkit focused on session visibility, recording and diagnostic context for AI-assisted workflows.
We'll share the supported workflows and availability as the product takes shape. No release date is announced yet.
inVtero.net / Deployment breadth
inVtero.net / Nested
Recursive introspection accounts for nested page-table and hypervisor relationships, rather than treating every address as if it belonged to one flat machine.
This is a logical view of nesting, not a claim that every platform exposes the same hardware page-table walk.
Observation
Recursive translation context
Reconstruct
Verify
Interpret
XAMMY / From generation to evidence
XAMMY / Generate
XAMMY explores watermarking at generation time. A configured sampler shapes token choices while the reader sees ordinary text.
MLX on Apple Silicon · vLLM on NVIDIA CUDA
Our engineers found that the proposed design reduces latency without increasing cost.
Selective AI-data reuse
RootCompute is designed to provide source and workload evidence to independent quality, diversity, grounding and eligibility checks. The goal is not to accept or reject synthetic data wholesale, but to make reuse decisions with more context.
This is a development direction, not a demonstrated training-performance claim. Memory presence does not by itself establish training consumption, and provenance does not establish factual truth.
The AI direction / inVtero.net roadmap
inVtero's next step is to make reconstructed state and integrity results usable by AI-assisted security systems. Models can help explain and prioritize changes; their hypotheses remain separate from deterministic verification.
01
Keep the observation source and its access boundaries explicit.
02
Recover address spaces, executable pages and topology context.
03
Compare evidence with reference material and preserve the result.
04
Use evidence to guide investigation, policy and future AI-assisted analysis.
Design partners & technical collaboration
Discuss RC RootCompute, inVtero and its add-ons, XAMMY evaluation, olol's next release or the forthcoming MCP-PTZ toolkit.
